CN117094975A - Method and device for detecting surface defects of steel and electronic equipment - Google Patents

Abstract

The invention provides a steel surface defect detection method, device and electronic equipment. The method includes: generating a binary segmentation threshold of the steel surface image to be tested based on the mean foreground grayscale and background grayscale of the steel surface image to be tested; sequentially performing grayscale equalization processing, filtering processing, and Edge enhancement processing is performed to obtain the first image; the first image is binarized using a binary segmentation threshold to obtain the second image; defects of the steel surface image to be tested are generated based on the second image and the pre-trained defect detection model Detection results; among them, the defect detection model is trained based on a preset training set. The preset training set includes multiple steel surface defect images and steel surface defect-free images. The steel surface defect images correspond to defective types. The invention can solve the problem that existing defect detection methods cannot meet the high-speed requirements for steel surface defect detection in actual industrial production.

Description

Steel surface defect detection methods, devices and electronic equipment

Technical field

The present invention relates to the technical field of steel detection, and in particular to a steel surface defect detection method, device and electronic equipment.

Background technique

As an important industrial raw material, steel is widely used in production and preparation scenarios such as aerospace equipment, automotive components, and industrial equipment. Therefore, the quality of steel directly determines the quality of the industrial cost of using it as raw materials. In order to ensure the quality of steel, the steel surface is generally tested for defects, and the quality of the steel is evaluated based on the degree of defects on the steel surface.

At present, manual inspection is usually used to detect steel surface defects. However, manual sampling cannot meet the high-speed requirements for steel surface defect detection in actual industrial production.

Contents of the invention

Embodiments of the present invention provide a steel surface defect detection method, device and electronic equipment to solve the problem that existing defect detection methods cannot meet the high-speed requirements for steel surface defect detection in actual industrial production.

In a first aspect, embodiments of the present invention provide a steel surface defect detection method, including:

According to the mean foreground gray value and background gray value of the steel surface image to be tested, the binary segmentation threshold of the steel surface image to be tested is generated;

The image of the steel surface to be tested is subjected to grayscale equalization, filtering and edge enhancement in sequence to obtain the first image;

Binarize the first image using a binary segmentation threshold to obtain the second image;

According to the second image and the pre-trained defect detection model, a defect detection result of the steel surface image to be tested is generated; wherein, the defect detection model is trained based on a preset training set, and the preset training set includes multiple steel surface defect images and Images of steel surfaces without defects, and images of steel surface defects corresponding to defective types.

In a possible implementation, a binary segmentation threshold of the steel surface image to be tested is generated based on the foreground gray average and the background gray average of the steel surface image to be tested, including:

Perform foreground and background segmentation operations on the steel surface image to be tested to obtain the foreground area and background area;

Calculate the average gray value of the steel surface image to be tested, the first pixel ratio of the foreground area and the steel surface image to be tested, and the second pixel ratio of the background area to the steel surface image to be tested;

According to the average gray value and the first pixel ratio, the average foreground gray value is obtained;

According to the average gray value and the second pixel ratio, the background gray mean value is obtained;

Calculate the average value of the foreground gray value and the background gray value, and define the average value as the binary segmentation threshold.

In a possible implementation, the first image is binarized using a binary segmentation threshold, including:

Set the grayscale value corresponding to the pixel point in the first image that is greater than the binary segmentation threshold as the first grayscale value;

The grayscale value corresponding to the pixel point in the first image that is smaller than the binary segmentation threshold is set as the second grayscale value.

In a possible implementation, the defect detection results of the surface image of the steel to be tested are generated based on the second image and the pre-trained defect detection model, including:

Perform closed operation processing on the second image to obtain a closed operation image;

Identify defective areas on closed operation images;

When there is a defective area in the closed operation image with a size larger than the preset threshold, the image area corresponding to the defective area in the closed operation image with a size larger than the preset threshold is marked in the steel surface image to be tested, and the marked area is The surface image of the steel to be tested is input into the defect detection model, and the defect detection results of the defect type corresponding to the surface image of the steel to be tested are obtained;

When there is no defect area with a size larger than the preset threshold in the closed operation image, a defect detection result is generated that the surface image of the steel to be tested is defect-free.

In a possible implementation, the image of the steel surface to be tested is sequentially subjected to grayscale equalization processing, filtering processing and edge enhancement processing to obtain the first image, including:

Obtain the grayscale mapping relationship for grayscale equalization processing based on the grayscale histogram of the steel surface image to be tested, and perform grayscale equalization processing on the steel surface image to be tested based on the grayscale mapping relationship to obtain the third image;

According to the spatial proximity of the third image and the pixel value similarity of the third image, the convolution weight used for filtering is obtained, and the third image is filtered according to the convolution weight to obtain a fourth image;

The horizontal gradient image and the vertical gradient image of the fourth image used for edge enhancement processing are acquired, and the first image is obtained based on the horizontal gradient image and the vertical gradient image.

In one possible implementation, a grayscale mapping relationship for grayscale equalization processing is obtained based on the grayscale histogram of the steel surface image to be tested, and the grayscale equalization is performed on the steel surface image to be tested based on the grayscale mapping relationship. Process to obtain the third image, including:

Divide the surface image of the steel to be tested into several partial images, and obtain the grayscale histograms corresponding to all partial images; among them, each partial image corresponds to a grayscale histogram, and the grayscale histogram is each grayscale level in the partial image. For the corresponding probability density function, the abscissa of the gray histogram is the gray level of each pixel in the local image, and the ordinate of the gray histogram is the frequency of occurrence of each gray level pixel in the local image;

Perform contrast limitation processing on the grayscale histogram to obtain a second grayscale histogram;

Obtain the cumulative probability density function corresponding to the second grayscale histogram, calculate the cumulative probability density function and the total number of grayscale levels, and obtain the grayscale mapping relationship for grayscale equalization processing; wherein, the cumulative probability density function is the second grayscale histogram. The integral of the second probability density function corresponding to the grayscale histogram;

Perform grayscale equalization processing on the local image according to the grayscale mapping relationship until all local image processing is completed;

All processed partial images are integrated to obtain a third image.

In a possible implementation, the convolution weight used for filtering is obtained according to the spatial proximity of the third image and the pixel value similarity of the third image, and the third image is filtered according to the convolution weight. After processing, the fourth image is obtained, including:

Define a center point in the third image;

Calculate the spatial proximity between each pixel in the third image and the center point, and the pixel value similarity between each pixel and the center point;

Multiply the spatial proximity of each pixel and the pixel value similarity to obtain the convolution weight of each pixel;

The fourth image is obtained based on the convolution weight of each pixel and the pixel value of each pixel.

In a possible implementation, obtaining the horizontal gradient image and the vertical gradient image of the fourth image used for edge enhancement processing, and obtaining the first image based on the horizontal gradient image and the vertical gradient image includes:

Obtain the horizontal matrix corresponding to the fourth image in the horizontal direction, and the vertical matrix corresponding to the fourth image in the vertical direction;

Perform planar convolution calculation on the horizontal matrix and the fourth image to obtain the horizontal gradient image;

Perform planar convolution calculation on the vertical matrix and the fourth image to obtain a vertical gradient image;

Perform a bitwise OR operation on the horizontal gradient image and the vertical gradient image to obtain the first image.

In a second aspect, embodiments of the present invention provide a steel surface defect detection device, including:

The generation module is used to generate the binary segmentation threshold of the steel surface image to be tested based on the foreground grayscale mean value and the background grayscale mean value of the steel surface image to be tested;

The first processing module is used to sequentially perform grayscale equalization processing, filtering processing and edge enhancement processing on the steel surface image to be tested to obtain the first image;

The second processing module is used to perform binarization processing on the first image using the binarization segmentation threshold to obtain the second image;

The detection module is used to generate defect detection results of the steel surface image to be tested based on the second image and a pre-trained defect detection model; wherein the defect detection model is trained based on a preset training set, and the preset training set includes multiple images The steel surface defect image and the steel surface defect-free image, the steel surface defect image corresponds to the defect type.

In a third aspect, embodiments of the present invention provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program Implement the steps of the method described in the above first aspect or any possible implementation of the first aspect.

Embodiments of the present invention provide a steel surface defect detection method, device and electronic equipment, which first generates a binary value of the steel surface image to be tested based on the foreground gray average and background gray average of the steel surface image to be tested that are obtained in advance segmentation threshold, and then perform pre-processing such as gray equalization, filtering, and edge enhancement on the steel surface image to be tested in order to obtain the first image, and then use the obtained binary segmentation threshold to binary the first image. processing to obtain a second image with only black and white pixels. Finally, based on the second image and the pre-trained defect detection model, the defect detection results of the steel surface image to be tested are obtained.

In this way, by binary processing the steel surface image to be measured through an appropriate binary segmentation threshold, the pixels in the first image can be converted into black or white, thereby enhancing the contrast and clarity of the first image to achieve Shorten the defect location identification time and accelerate the technical effect of the defect detection process. And since there are only two colors in the second image, and the small number of colors in the second image also illustrates the small amount of data in the second image, the speed of subsequent image processing will be significantly faster than that of the image that has not been binarized. . In addition, the detection results obtained by using a defect detection model trained based on a large amount of data are also more accurate. Therefore, the present invention can detect defects on the steel surface at high speed and accurately in actual industrial production.

Description of the drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings in the following description are only illustrative of the present invention. For some embodiments, for those of ordinary skill in the art, other drawings can be obtained based on these drawings without exerting creative efforts.

Figure 1 is an implementation flow chart of a steel surface defect detection method provided by an embodiment of the present invention;

Figure 2 is a schematic diagram of a grayscale histogram of a partial image provided by an embodiment of the present invention;

Figure 3 is a schematic diagram of a grayscale threshold setting and updating a grayscale histogram provided by an embodiment of the present invention;

Figure 4 is a schematic comparison diagram before and after an image morphology closed operation process provided by an embodiment of the present invention;

Figure 5 is a schematic structural diagram of a binary matrix provided by an embodiment of the present invention;

Figure 6 is a specific implementation flow chart of a steel surface defect detection method provided by an embodiment of the present invention;

Figure 7 is a schematic structural diagram of a steel surface defect detection device provided by an embodiment of the present invention;

Figure 8 is a schematic diagram of an electronic device provided by an embodiment of the present invention.

Detailed ways

In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the present invention in unnecessary detail.

In order to make the purpose, technical solutions and advantages of the present invention clearer, specific embodiments will be described below in conjunction with the accompanying drawings.

The surface quality of steel is one of the most important quality factors of steel. The surface quality of steel directly affects the performance and quality of its final product. However, during steel processing, different types of defects such as scars, cracks, scratches, holes or pitting may appear on the steel surface due to raw materials or processes. These defects not only affect the appearance of the product, but also reduce the corrosion resistance, wear resistance, fatigue strength and other properties of the product. Therefore, how to detect surface defects of steel to improve the surface quality of steel is a matter of great concern to steel processing enterprises.

However, the existing technology uses manual spot inspection to detect defects on the steel surface, which not only cannot meet the high-speed requirements for detecting steel surface defects in actual industrial production, but also may cause defects to be missed during the detection process. It can be seen that the existing technology cannot meet the high-precision defect detection in high-speed scenarios. Therefore, the present invention provides a high-precision steel surface defect detection method that can solve the problems of the existing technology and is suitable for high-speed scenarios.

Figure 1 is a flow chart of the implementation of the steel surface defect detection method provided by the embodiment of the present invention. The details are as follows:

Step 101: Generate a binary segmentation threshold of the steel surface image to be tested based on the mean foreground gray value and background gray value of the steel surface image to be tested.

In some embodiments, obtaining the surface image of the steel to be measured can be obtained by 1. dividing the surface of the steel. 2. Use an ordinary camera to shoot each divided area. 3. Integrate all the acquired images to obtain the surface image of the steel to be tested. In addition to the above methods, the surface image of the steel to be tested can also be obtained through a line scan camera.

It should be noted that the principle of a line scan camera is that the camera and the object being photographed move relatively uniformly, thereby obtaining a complete and clear image of the object being photographed. Due to the simple structure, low cost and high flexibility of the line scan camera, the present invention preferentially uses the line scan camera to obtain the surface image of the steel to be measured.

In some embodiments, a foreground and background segmentation model can be used to perform foreground and background segmentation on the steel surface image to be measured, to obtain the foreground area and background area of the steel surface image to be measured, and then calculate the foreground grayscale mean and background grayscale mean.

In some embodiments, multiple existing steel surface images, the foreground area of the steel surface image, and the background area of the steel surface image can be used to train the foreground and background segmentation model to ensure the accuracy of the foreground and background segmentation model.

In a possible implementation, the present invention does not limit the type of algorithm used in the foreground and background segmentation model. For example, a fully convolutional neural network can be used as the algorithm framework of the foreground and background segmentation model.

In some embodiments, after the steel surface image to be measured is divided into a foreground area and a background area, the total average gray value of the entire image of the steel surface image to be measured can be used, as well as the proportion of the foreground area, background area and the entire image. The ratio relationship is used to obtain the mean foreground gray level and the mean background gray level. Its specific implementation can be:

1. Calculate the total average gray value of the entire image of the steel surface image to be tested.

2. Calculate the first pixel ratio of the pixels in the foreground area to the total pixels.

3. Calculate the second pixel ratio of the pixels in the background area to the total pixels.

4. Multiply the total average gray value by the first pixel ratio and the second pixel ratio respectively to obtain the foreground gray mean and background gray mean of the steel surface image to be tested.

In some embodiments, the binary segmentation threshold may be the average of the foreground grayscale mean and the background grayscale mean.

In some embodiments, obtaining the binary segmentation threshold can provide parameter data for subsequent binarization processing, thereby facilitating subsequent image processing and improving the defect detection speed of the steel surface image to be tested.

Specifically, the specific implementation steps of step 101 may include:

1. Train the foreground and background segmentation model to obtain the trained foreground and background segmentation model.

2. Use a line scan camera to obtain the surface image of the steel to be tested.

3. Use the foreground and background segmentation model to segment the steel surface image to be tested into a foreground area and a background area.

4. Calculate the total average gray value of the steel surface image to be tested, calculate the first pixel ratio of pixels in the foreground area to the total pixels, and the second pixel ratio of the pixels in the background area to the total pixels. For example, assuming that the pixel points in the foreground area are 50, the pixel points in the background area are 100, and the total pixel points are 200, then the first pixel ratio is 0.25 and the second pixel ratio is 0.5.

5. Multiply the first pixel ratio and the total average gray value to obtain the foreground gray average, and multiply the second pixel ratio and the total average gray to obtain the background gray average. For example, assuming that the overall average gray value is 50, the average foreground gray value is 12.5 and the average background gray value is 25.

6. Calculate the average of the foreground gray average and the background gray average, and use this average as the binary segmentation threshold. For example, according to the above calculation, the binary segmentation threshold is That is 18.75.

Step 102, perform grayscale equalization processing, filtering processing and edge enhancement processing in sequence on the steel surface image to be tested to obtain the first image.

Specifically, the main technical feature of the grayscale equalization process provided by the present invention is to transform the histogram distribution of an image into an approximately uniform distribution, thereby increasing the contrast of the image.

First, the grayscale equalization process is introduced. Specifically, the specific implementation steps of the gray level equalization process may be steps 201-204:

Step 201: Divide the surface image of the steel to be tested into several partial images, and obtain the grayscale histograms corresponding to all partial images.

In some embodiments, the surface image of the steel to be measured is divided into several partial images, and each partial image can be processed accordingly to obtain a more accurate and beautiful grayscale equalized image.

In some embodiments, the present invention does not limit the specific size of the partial image, but the size of the divided partial image should be in a moderate range and should not be too small or too large.

In some embodiments, the grayscale histogram reflects the relationship between the frequency of occurrence of each grayscale pixel point in an image and the grayscale level. Therefore, when applied to the present invention, the abscissa of the gray histogram can be the gray level of each pixel point in the local image, and the ordinate of the gray histogram can be the pixel point of each gray level appearing in the local image. frequency, and each local image corresponds to a grayscale histogram.

It should be noted that the grayscale histogram can also be regarded as the probability density function corresponding to each grayscale level in the local image.

Therefore, it can be based on Calculate the probability density corresponding to each gray level. In the formula, P represents the probability density function, i represents the current gray level, n _i represents the number of pixels whose gray level is the current gray level, L represents the total number of gray levels, and n represents the total number of pixels.

In order to express the probability of the grayscale histogram more accurately, the following uses a specific embodiment to explain the grayscale histogram:

In some embodiments, it is assumed that there are 36 pixels in a certain local image, among which there are 6 pixels with gray level 1, 4 pixels with gray level 2, and 3 pixels. There are 6 pixels, 6 pixels with gray level 4, 2 pixels with gray level 5, and 12 pixels with gray level 6. It can be seen that the frequency of pixels with gray level 1 is The frequency of pixels with gray level 2 is/> The frequency of pixels with gray level 3 is/> The frequency of pixels with gray level 4 is/> The frequency of pixels with gray level 5 is/> The frequency of pixels with gray level 6 is/> According to the frequency corresponding to each gray level, the gray histogram of the local image under the current hypothesis is obtained as shown in Figure 2.

Step 202: Perform contrast limitation processing on the grayscale histogram to obtain a second grayscale histogram.

In some embodiments, as shown in Figure 3, contrast limitation processing can be understood as setting a grayscale threshold in the grayscale histogram, and when there are pixels with a probability density greater than the grayscale threshold in the grayscale histogram, Pixels are evenly distributed to each gray level. Then, the grayscale histogram is updated based on the frequency of occurrence of the assigned pixels of each grayscale level in the local image to obtain a second grayscale histogram.

In one possible implementation, by redistributing the pixels of the local image, the number of pixels within a certain gray level range can be made approximately the same, thereby facilitating subsequent equalization operations.

It should be noted that the present invention does not limit the specific value of the grayscale threshold, for example, it can be 25%, 30% or 27%.

Step 203: Obtain the cumulative probability density function corresponding to the second grayscale histogram, calculate the cumulative probability density function and the total number of grayscale levels, and obtain a grayscale mapping relationship for grayscale equalization processing.

In some embodiments, the cumulative probability density function is the integral of the second probability density function corresponding to the second grayscale histogram.

It should be noted that after obtaining the cumulative probability density function corresponding to the second grayscale histogram, the above cumulative probability density function also needs to be normalized to the range of (0, L-1).

In a possible implementation, by normalizing the cumulative probability density function to the range of (0, L-1), the second histogram can be changed from a certain gray level interval in the comparison set to one in all gray levels. Uniform distribution within the range of degree levels.

In some embodiments, the grayscale mapping relationship can be obtained by multiplying the cumulative probability density function and the total number of grayscale levels.

Step 204: perform grayscale equalization processing on the local image according to the grayscale mapping relationship until all local image processing is completed.

In some embodiments, step 201 has stated that each partial image corresponds to a grayscale histogram, and therefore, each partial image also corresponds to a grayscale mapping relationship. It can be seen that the grayscale mapping relationship corresponding to each local image may be different. Therefore, in practical applications, the partial image needs to be calculated on its corresponding grayscale mapping relationship to obtain the processed partial image.

Step 205: Integrate all processed partial images to obtain a third image.

In some embodiments, all processed partial images are integrated according to the division rules in step 201, that is, a grayscale equalized image of the steel surface to be tested is obtained.

The foregoing content introduces the grayscale equalization process, and the filtering process is introduced below.

In some embodiments, due to imperfections in transmission media and recording equipment, images are often contaminated by various noises during their transmission and recording process. Therefore, it is necessary to filter the image during image processing. It should be noted that filtering processing can also be regarded as denoising processing, which is a technical means to denoise and smooth images.

Specifically, the specific implementation steps of the filtering process may be steps 301-304:

Step 301: Define a center point in the third image.

In some embodiments, the present invention does not limit the specific selection position of the center point, but the center point should be selected from a moderately located area in the third image and should not be excessively biased in a certain direction.

Step 302: Calculate the spatial proximity between each pixel in the third image and the center point, and the pixel value similarity between each pixel and the center point.

In some embodiments, it can be based on Calculate the spatial proximity of each pixel to the center point, and the pixel value similarity between each pixel and the center point. In the formula, w _s represents the spatial proximity, w _r represents the pixel value similarity, (k, l) represents the pixel point, (x, y) represents the center point, f (k, l) represents the pixel value of the pixel point, f (x,y) represents the pixel value of the center point, and e represents the base of the natural logarithm.

Step 303: Multiply the spatial proximity of each pixel by the pixel value similarity to obtain the convolution weight of each pixel;

In some embodiments, the convolution of each pixel can be obtained according to w (x, y, k, l) = w _s (x, y, k, l) × w _r (x, y, k, l) weight. In the formula, w represents the convolution weight.

Step 304: Obtain a fourth image based on the convolution weight of each pixel and the pixel value of each pixel.

In some embodiments, the fourth image can be obtained through the following steps: 1. Calculate the product of the pixel values of all pixels in the third image and the convolution weight, and add all the products to obtain the sum of products; 2. Calculate all The sum of the convolution weights; 3. Compare the sum of the products with the sum of the convolution weights to obtain the fourth image.

Specifically, the formula for the above steps can be: In the formula, g(x,y) represents the fourth image, and S(x,y) represents the area within the range of (2N+1)*(2N+1) centered on (x,y).

It should be noted that the above S(x,y) may refer to the area where the third image is located in the present invention.

In some embodiments, since the fourth image after filtering may be relatively smooth, the edges of the defects are not strong. Therefore, edge enhancement processing and filtering processing can be combined to achieve the technical effects of removing noise in the image and enhancing the edges of defective areas.

The foregoing content introduces the filtering process, and the edge enhancement process is introduced below.

In some embodiments, the main technical content of the edge enhancement processing provided by the present invention can be to use operators to perform convolution calculations in the horizontal and vertical directions of the image to obtain gradient results in two directions. The gradient results are combined to achieve edge enhancement of defective areas.

Specifically, the specific implementation steps of edge enhancement processing may be steps 401-404:

Step 401: Obtain the horizontal matrix corresponding to the fourth image in the horizontal direction and the vertical matrix corresponding to the fourth image in the vertical direction.

In some embodiments, since the edge enhancement process uses an operator to perform convolution calculation on the image, the internal parameters of the horizontal matrix and the vertical matrix corresponding to the fourth image are determined by the type of the operator.

In some embodiments, taking the Sobel operator as an example, the horizontal matrix corresponding to the fourth image can be The vertical matrix corresponding to the fourth image can be/>

Step 402: Perform planar convolution calculation on the horizontal matrix and the fourth image to obtain a horizontal gradient image.

In some embodiments, it can be based on Get a horizontal gradient image. In the formula, X(x,y) represents the horizontal gradient image.

In some embodiments, performing planar convolution calculation on the horizontal matrix and the fourth image is equivalent to using the horizontal matrix to extract edges in the horizontal direction.

Step 403: Perform planar convolution calculation on the vertical matrix and the fourth image to obtain a vertical gradient image.

In some embodiments, it can be based on Get a vertical gradient image. In the formula, Y(x,y) represents the vertical gradient image.

In some embodiments, performing planar convolution calculation on the vertical matrix and the fourth image is equivalent to using the vertical matrix to extract edges in the vertical direction.

Step 404: Perform a bitwise OR operation on the horizontal gradient image and the vertical gradient image to obtain the first image.

In some embodiments, it can be based on Get the first image. In the formula, G(x,y) represents the first image,/> Represents bitwise OR operation.

In some embodiments, the edge information of the entire image can be obtained by combining the edge information in the horizontal and vertical directions, highlighting the boundary between the defective area and the background, making subsequent positioning and marking of the defective area simpler.

Step 103: Binarize the first image using a binary segmentation threshold to obtain a second image.

In some embodiments, the specific steps of the binarization process may be: 1. Set the grayscale value corresponding to the pixel point in the first image that is greater than the binary segmentation threshold as the first grayscale value. 2. Set the gray value corresponding to the pixel point in the first image that is smaller than the binary segmentation threshold as the second gray value.

It should be noted that, in order to make the contrast of the first image after binarization processing higher, and with reference to the value range of the gray value, the value of the first gray value can be 255, and the value of the second gray value can be 255. The value can be 0.

In a possible implementation, binarizing the first image can convert the pixels in the first image to black or white, thereby enhancing the contrast and clarity of the first image. In addition, since there are only two colors in the second image, the data amount of the second image will also be reduced compared to the previous image.

It is worth mentioning that the small amount of data in the second image also indicates that the speed of subsequent image processing will be faster than that of images that are not binarized. Therefore, performing binarization processing on the first image can achieve the technical effect of the present invention being suitable for high-speed scenes.

Step 104: Generate defect detection results of the steel surface image to be tested based on the second image and the pre-trained defect detection model; wherein the defect detection model is trained based on a preset training set, and the preset training set includes multiple steel surfaces. The defect image and the steel surface defect-free image, the steel surface defect image corresponds to the defect type.

The foregoing content introduces edge enhancement processing, and the morphological closing operation processing is introduced below.

In some embodiments, in order to make the image input into the defect detection model more accurate, the second image may also be subjected to morphological closed operation processing to obtain a closed operation image. As shown in Figure 4, the comparison diagram before and after the image morphological closing operation is processed. Picture a is the image before the morphological closing operation. Picture b is the image after the morphological closing operation. The white parts in pictures a and b are is the defective area, and the black part is the defect-free area. It can be seen that performing morphological closing operation processing can not only remove small noise in the background of the second image, but also fill small holes in connected areas and disconnect adjacent objects with fewer connections.

Specifically, the specific implementation steps of the morphological closing operation processing may be steps 501-503:

Step 501: Define a binary matrix composed of structural elements. As shown in Figure 5, the structural diagram of a binary matrix is shown. The black part in the figure represents the effective part of the structural element, and the value is 1. The white part in the figure represents the ineffective part of the structural element, and the value is 0.

Step 502: Use structural elements to expand the second image. That is, structural elements are used to move from each pixel point in the second image to convolve the second image.

It should be noted that for each pixel during the expansion operation, as long as the pixel value of at least one pixel in the portion where the structural element intersects with the surrounding pixels of a certain pixel is non-zero, the pixel will be The pixel value of the pixel is set to 1.

Step 503: Use structural elements to perform an erosion operation on the expanded second image. That is, the structural elements are used to move from each pixel point in the expanded second image, and the expanded second image is convolved.

It should be noted that for each pixel during the erosion operation, only when the pixel values of all pixels in the intersection between the adjacent area of a pixel and the structural element are non-zero, the pixel can be converted into a pixel. The pixel value of the pixel is set to 1; otherwise, the pixel value of the pixel is set to 0.

In some embodiments, the training process of the defect detection model may be: 1. Train the defect detection model based on a preset training set. The preset training set includes multiple steel surface defect images and steel surface defect-free images, as well as defect types corresponding to the steel surface defect images. 2. Adjust and optimize the parameters of the defect detection model based on the preset training set and application scenarios to obtain the trained defect detection model.

It should be noted that the defect detection model can be a deep learning model, and when the training time is insufficient, transfer learning can also be used to shorten the training process of the defect detection model.

It should be noted that assuming that the defect detection model is a deep learning model, during the actual training process, due to the large differences in the sizes of individual defect areas, in order to improve the feature extraction capabilities of the defect detection model for defect areas of various sizes, you can The neck structure of the deep learning model, that is, the feature pyramid network, is replaced with the Inception block.

In some embodiments, after obtaining the closed operation image, defective area identification can be performed on the closed operation image.

Specifically, when there is no defect area with a size larger than a preset threshold in the closed operation image, and the similarity between the steel surface image to be tested corresponding to the closed operation image and the defect-free steel surface image is greater than the first preset value, it is determined The surface image of the steel to be tested corresponding to the closed operation image has no defects.

In a possible implementation, the present invention does not limit the value of the first preset value. For example, the first preset value may be 90%, 95% or 96%, etc.

In some embodiments, when there is a defective area with a size larger than a preset threshold in the closed operation image, the location of the defective area in the closed operation image can be marked on the steel surface image to be tested by referring to the position of the defective area, and the mark can be The final surface image of the steel to be tested is saved to the suspicious image queue, and then input into the defect detection model to obtain the defect detection results of the defect type corresponding to the surface image of the steel to be tested.

In one possible implementation, marking the defect area in the original image and then inputting it into the defect detection model can not only achieve preliminary positioning of the location of the defect, save defect positioning time during the detection process, but also ensure safety from the source. The accuracy of images and defect results solves the problem of detection errors that may be caused by inputting processed images into the defect detection model.

Specifically, when there is a defect area with a size larger than a preset threshold in the closed operation image, and the marked defect area is a steel surface defect image whose similarity is greater than the second preset value, the steel surface to be tested corresponding to the closed operation image is generated. Defect detection results of the defect type corresponding to the image.

In a possible implementation, the present invention does not limit the value of the second preset value. For example, the second preset value may be 90%, 95%, 96%, etc.

In some embodiments, types of defects may include cracks, scratches, folds, ears, scabs, rust, bubbles, pitting, looseness, non-metallic inclusions, burns, white spots, weld scars, or end burrs. wait.

It should be noted that when there is a defect area with a size larger than the preset threshold in the closed operation image, and no steel surface defect image similar to the marked defect area is found in the steel surface defect image, it is determined that the marked defect area is not Defects, and no defect types.

It should be noted that there may be multiple defective areas in the surface image of the steel to be tested. When the marked surface image of the steel to be tested is input into the defect detection model for defect detection, the defect detection model detects defects in multiple defective areas at the same time. Detected. Therefore, the present invention does not need to detect defects in defective areas one by one. It can be seen that the detection process of the present invention can meet the high-speed requirements for steel surface defect detection in actual industrial production.

It should be noted that after it is determined that the marked surface image of the steel to be tested is defective, the surface image of the steel to be tested can be stored in a preset training set, and new training data can be added to the preset training set.

In some embodiments, in practical applications, monitoring can be set up in the suspicious image queue. If a marked steel surface image to be tested is added to the suspicious image queue, the marked steel surface image to be tested is immediately input to the defect detection Detection is carried out in the model and the detection results of the surface image of the steel to be tested are obtained.

It is worth mentioning that when monitoring the suspicious image queue is set up to ensure that there are no other images to be tested in the suspicious image queue, the new images to be tested in the suspicious image queue can be input into the defect detection model as soon as possible to solve the problem. The problem of wasting time is caused by the unresponsiveness of the image to be tested after being added to the suspicious image queue.

As shown in the specific implementation flow chart of the steel surface defect detection method in Figure 6, it can be seen that the present invention can take into account the processing speed requirements and defect detection accuracy requirements in high-speed steel defect detection tasks.

Embodiments of the present invention provide a steel surface defect detection method, which first generates a binary segmentation threshold of the steel surface image to be tested based on the foreground gray average and background gray average of the steel surface image to be tested, and then The first image is obtained by sequentially performing preprocessing such as gray equalization, filtering, and edge enhancement on the steel surface image to be tested, and then binarizes the first image through the obtained binary segmentation threshold to obtain pixels. There is only a second image in black and white. Finally, based on the second image and the pre-trained defect detection model, the defect detection results of the surface image of the steel to be tested are obtained.

In this way, by binary processing the steel surface image to be measured through an appropriate binary segmentation threshold, the pixels in the first image can be converted into black or white, thereby enhancing the contrast and clarity of the first image to achieve Shorten the defect location identification time and accelerate the technical effect of the defect detection process. And since there are only two colors in the second image, and the small number of colors in the second image also illustrates the small amount of data in the second image, the speed of subsequent image processing will be significantly faster than that of the image that has not been binarized. . In addition, the detection results obtained by using a defect detection model trained based on a large amount of data are also more accurate. Therefore, the present invention can detect defects on the steel surface at high speed and accurately in actual industrial production.

It should be understood that the sequence number of each step in the above embodiment does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

The following are device embodiments of the present invention. For details that are not described in detail, reference may be made to the above corresponding method embodiments.

Figure 7 shows a schematic structural diagram of a steel surface defect detection device provided by an embodiment of the present invention. For convenience of explanation, only the parts related to the embodiment of the present invention are shown. The details are as follows:

As shown in Figure 7, the steel surface defect detection device 7 includes:

The generation module 71 is used to generate a binary segmentation threshold of the steel surface image to be tested based on the foreground grayscale mean value and the background grayscale mean value of the steel surface image to be tested;

The first processing module 72 is used to sequentially perform grayscale equalization processing, filtering processing and edge enhancement processing on the steel surface image to be tested to obtain the first image;

The second processing module 73 is used to binarize the first image using the binarization segmentation threshold to obtain the second image;

The detection module 74 is used to generate defect detection results of the steel surface image to be tested based on the second image and a pre-trained defect detection model; wherein the defect detection model is trained based on a preset training set, and the preset training set includes multiple There are images of steel surface defects and images of steel surfaces without defects, and the steel surface defect images correspond to defective types.

In a possible implementation, the generation module 71 is specifically used to:

Perform foreground and background segmentation operations on the steel surface image to be tested to obtain the foreground area and background area;

Calculate the average gray value of the steel surface image to be tested, the first pixel ratio of the foreground area and the steel surface image to be tested, and the second pixel ratio of the background area to the steel surface image to be tested;

According to the average gray value and the first pixel ratio, the average foreground gray value is obtained;

According to the average gray value and the second pixel ratio, the background gray mean value is obtained;

Calculate the average value of the foreground gray value and the background gray value, and define the average value as the binary segmentation threshold.

In a possible implementation, the second processing module 73 is specifically used to:

Set the grayscale value corresponding to the pixel point in the first image that is greater than the binary segmentation threshold as the first grayscale value;

The grayscale value corresponding to the pixel point in the first image that is smaller than the binary segmentation threshold is set as the second grayscale value.

In a possible implementation, the detection module 74 is specifically used to:

Perform closed operation processing on the second image to obtain a closed operation image;

Identify defective areas on closed operation images;

When there is a defective area in the closed operation image with a size larger than the preset threshold, the image area corresponding to the defective area in the closed operation image with a size larger than the preset threshold is marked in the steel surface image to be tested, and the marked area is The surface image of the steel to be tested is input into the defect detection model, and the defect detection results of the defect type corresponding to the surface image of the steel to be tested are obtained;

When there is no defect area with a size larger than the preset threshold in the closed operation image, a defect detection result is generated that the surface image of the steel to be tested is defect-free.

In a possible implementation, the first processing module 72 is specifically used to:

Obtain the grayscale mapping relationship for grayscale equalization processing based on the grayscale histogram of the steel surface image to be tested, and perform grayscale equalization processing on the steel surface image to be tested based on the grayscale mapping relationship to obtain the third image;

According to the spatial proximity of the third image and the pixel value similarity of the third image, the convolution weight used for filtering is obtained, and the third image is filtered according to the convolution weight to obtain a fourth image;

The horizontal gradient image and the vertical gradient image of the fourth image used for edge enhancement processing are acquired, and the first image is obtained based on the horizontal gradient image and the vertical gradient image.

In a possible implementation, the first processing module 72 is specifically used to:

Divide the surface image of the steel to be tested into several partial images, and obtain the grayscale histograms corresponding to all partial images; among them, each partial image corresponds to a grayscale histogram, and the grayscale histogram is each grayscale level in the partial image. For the corresponding probability density function, the abscissa of the gray histogram is the gray level of each pixel in the local image, and the ordinate of the gray histogram is the frequency of occurrence of each gray level pixel in the local image;

Perform contrast limitation processing on the grayscale histogram to obtain a second grayscale histogram;

Obtain the cumulative probability density function corresponding to the second grayscale histogram, calculate the cumulative probability density function and the total number of grayscale levels, and obtain the grayscale mapping relationship for grayscale equalization processing; wherein, the cumulative probability density function is the second grayscale histogram. The integral of the second probability density function corresponding to the grayscale histogram;

Perform grayscale equalization processing on the local image according to the grayscale mapping relationship until all local image processing is completed;

All processed partial images are integrated to obtain a third image.

In a possible implementation, the first processing module 72 is specifically used to:

Define a center point in the third image;

Calculate the spatial proximity between each pixel in the third image and the center point, and the pixel value similarity between each pixel and the center point;

Multiply the spatial proximity of each pixel and the pixel value similarity to obtain the convolution weight of each pixel;

The fourth image is obtained based on the convolution weight of each pixel and the pixel value of each pixel.

In a possible implementation, the first processing module 72 is specifically used to:

Obtain the horizontal matrix corresponding to the fourth image in the horizontal direction, and the vertical matrix corresponding to the fourth image in the vertical direction;

Perform planar convolution calculation on the horizontal matrix and the fourth image to obtain the horizontal gradient image;

Perform planar convolution calculation on the vertical matrix and the fourth image to obtain a vertical gradient image;

Perform a bitwise OR operation on the horizontal gradient image and the vertical gradient image to obtain the first image.

Embodiments of the present invention provide a steel surface defect detection device, which first generates a binary segmentation threshold of the steel surface image to be measured based on the foreground gray average and background gray average of the steel surface image to be measured, and then The first image is obtained by sequentially performing preprocessing such as gray equalization, filtering, and edge enhancement on the steel surface image to be tested, and then binarizes the first image through the obtained binary segmentation threshold to obtain pixels. There is only a second image in black and white. Finally, based on the second image and the pre-trained defect detection model, the defect detection results of the surface image of the steel to be tested are obtained.

In this way, by binary processing the steel surface image to be measured through an appropriate binary segmentation threshold, the pixels in the first image can be converted into black or white, thereby enhancing the contrast and clarity of the first image to achieve Shorten the defect location identification time and accelerate the technical effect of the defect detection process. And since there are only two colors in the second image, and the small number of colors in the second image also illustrates the small amount of data in the second image, the speed of subsequent image processing will be significantly faster than that of the image that has not been binarized. . In addition, the detection results obtained by using a defect detection model trained based on a large amount of data are also more accurate. Therefore, the present invention can detect defects on the steel surface at high speed and accurately in actual industrial production.

Figure 8 is a schematic diagram of an electronic device provided by an embodiment of the present invention. As shown in FIG. 8 , the electronic device 8 of this embodiment includes: a processor 80 , a memory 81 , and a computer program 82 stored in the memory 81 and executable on the processor 80 . When the processor 80 executes the computer program 82, it implements the steps in each of the above embodiments of the steel surface defect detection method, such as steps 101 to 104 shown in FIG. 1 . Alternatively, when the processor 80 executes the computer program 82, it implements the functions of each module/unit in each of the above device embodiments, such as the functions of the modules/units 71 to 74 shown in FIG. 7 .

Exemplarily, the computer program 82 can be divided into one or more modules/units, the one or more modules/units are stored in the memory 81 and executed by the processor 80 to complete this invention. The one or more modules/units may be a series of computer program instruction segments capable of completing specific functions. The instruction segments are used to describe the execution process of the computer program 82 in the electronic device 8 . For example, the computer program 82 may be divided into modules/units 71 to 74 as shown in FIG. 7 .

The electronic device 8 may include, but is not limited to, a processor 80 and a memory 81 . Those skilled in the art can understand that FIG. 8 is only an example of the electronic device 8 and does not constitute a limitation of the electronic device 8. It may include more or fewer components than shown in the figure, or some components may be combined, or different components may be used. , for example, the electronic device may also include input and output devices, network access devices, buses, etc.

The so-called processor 80 can be a central processing unit (Central Processing Unit, CPU), or other general-purpose processor, digital signal processor (Digital Signal Processor, DSP), application specific integrated circuit (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

The memory 81 may be an internal storage unit of the electronic device 8 , such as a hard disk or memory of the electronic device 8 . The memory 81 may also be an external storage device of the electronic device 8, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), or a secure digital (SD) equipped on the electronic device 8. card, flash card, etc. Further, the memory 81 may also include both an internal storage unit of the electronic device 8 and an external storage device. The memory 81 is used to store the computer program and other programs and data required by the electronic device. The memory 81 can also be used to temporarily store data that has been output or is to be output.

Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above functional units and modules is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs. Module completion means dividing the internal structure of the device into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be hardware-based. It can also be implemented in the form of software functional units. In addition, the specific names of each functional unit and module are only for the convenience of distinguishing each other and are not used to limit the scope of protection of the present application. For the specific working processes of the units and modules in the above system, please refer to the corresponding processes in the foregoing method embodiments, and will not be described again here.

In the above embodiments, each embodiment is described with its own emphasis. For parts that are not detailed or documented in a certain embodiment, please refer to the relevant descriptions of other embodiments.

Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may implement the described functionality using different methods for each specific application, but such implementations should not be considered to be beyond the scope of the present invention.

In the embodiments provided by the present invention, it should be understood that the disclosed apparatus/electronic equipment and methods can be implemented in other ways. For example, the device/electronic device embodiments described above are only illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units. Or components can be combined or can be integrated into another system, or some features can be omitted, or not implemented. On the other hand, the coupling or direct coupling or communication connection between each other shown or discussed may be through some interfaces, indirect coupling or communication connection of devices or units, which may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or they may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware or software functional units.

If the integrated module/unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can implement all or part of the processes in the methods of the above embodiments, and can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can be stored in a computer-readable storage medium. When the program is executed by the processor, the steps of each of the above embodiments of the steel surface defect detection method can be implemented. Wherein, the computer program includes computer program code, which may be in the form of source code, object code, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (Read-Only Memory, ROM) , random access memory (Random Access Memory, RAM), electrical carrier signals, telecommunications signals, and software distribution media, etc.

The above-described embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still implement the above-mentioned implementations. The technical solutions described in the examples are modified, or some of the technical features are equivalently replaced; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of each embodiment of the present invention, and should be included in within the protection scope of the present invention.

Claims (10)

1. A method for detecting surface defects of steel, comprising:

generating a binarization segmentation threshold value of the steel surface image to be detected according to a front Jing Huidu mean value and a background gray level mean value of the steel surface image to be detected;

sequentially carrying out gray level equalization treatment, filtering treatment and edge enhancement treatment on the steel surface image to be detected to obtain a first image;

performing binarization processing on the first image by using the binarization segmentation threshold value to obtain a second image;

generating a defect detection result of the steel surface image to be detected according to the second image and a pre-trained defect detection model; the defect detection model is obtained by training based on a preset training set, the preset training set comprises a plurality of steel surface defect images and steel surface defect-free images, and the steel surface defect images are corresponding to defect types.

2. The method for detecting a steel surface defect according to claim 1, wherein the generating the binarized segmentation threshold of the steel surface image to be detected according to the front Jing Huidu mean value and the background gray-level mean value of the steel surface image to be detected comprises:

performing foreground and background segmentation operation on the steel surface image to be detected to obtain a foreground region and a background region;

calculating an average gray value of the steel surface image to be detected, a first pixel ratio of the foreground region to the steel surface image to be detected, and a second pixel ratio of the background region to the steel surface image to be detected;

obtaining the front Jing Huidu average value according to the average gray value and the first pixel ratio;

obtaining the background gray average value according to the average gray value and the second pixel ratio;

an average of the front Jing Huidu average and the background grayscale average is calculated and defined as the binarized segmentation threshold.

3. The method for detecting a surface defect of steel according to claim 1, wherein the binarizing the first image using the binarized segmentation threshold value comprises:

Setting a gray value corresponding to a pixel point larger than the binarization segmentation threshold value in the first image as a first gray value;

and setting the gray value corresponding to the pixel point smaller than the binarization segmentation threshold in the first image as a second gray value.

4. The method for detecting the defects of the steel surface according to claim 1, wherein the generating the defect detection result of the image of the steel surface to be detected according to the second image and a pre-trained defect detection model comprises:

performing a closed operation process on the second image to obtain a closed operation image;

performing defect area identification on the closed operation image;

when a defect area with the size larger than a preset threshold exists in the closed operation image, marking an image area corresponding to the defect area with the size larger than the preset threshold in the closed operation image in the steel surface image to be detected, and inputting the marked steel surface image to be detected into the defect detection model to obtain a defect detection result of a defect type corresponding to the steel surface image to be detected;

and when the defect area with the size larger than a preset threshold value does not exist in the closed operation image, generating a defect detection result that the steel surface image to be detected is defect-free.

5. The method for detecting a surface defect of steel according to claim 1, wherein the sequentially performing gray-scale equalization processing, filtering processing and edge enhancement processing on the steel surface image to be detected to obtain a first image comprises:

acquiring a gray mapping relation for gray level equalization processing according to the gray level histogram of the steel surface image to be detected, and carrying out gray level equalization processing on the steel surface image to be detected according to the gray level mapping relation to obtain a third image;

acquiring a convolution weight for filtering according to the spatial proximity of the third image and the pixel value similarity of the third image, and filtering the third image according to the convolution weight to obtain a fourth image;

acquiring a horizontal gradient image and a vertical gradient image of the fourth image for edge enhancement processing, and obtaining the first image according to the horizontal gradient image and the vertical gradient image.

6. The method for detecting a defect on a steel surface according to claim 5, wherein the step of obtaining a gray mapping relation for gray equalization according to a gray histogram of the steel surface image to be detected, and performing gray equalization on the steel surface image to be detected according to the gray mapping relation, to obtain a third image, comprises:

Dividing the steel surface image to be detected into a plurality of partial images, and obtaining gray histograms corresponding to all the partial images; each local image corresponds to a gray level histogram, the gray level histogram is a probability density function corresponding to each gray level in the local image, the abscissa of the gray level histogram is the gray level of each pixel point in the local image, and the ordinate of the gray level histogram is the frequency of occurrence of each gray level pixel point in the local image;

performing contrast limiting processing on the gray level histogram to obtain a second gray level histogram;

acquiring an accumulated probability density function corresponding to the second gray level histogram, and calculating the accumulated probability density function and the total gray level number to obtain a gray level mapping relation for gray level equalization processing; wherein the cumulative probability density function is an integral of a second probability density function corresponding to the second gray level histogram;

carrying out gray level equalization processing on the partial images according to the gray level mapping relation until all partial image processing is completed;

and integrating all the processed partial images to obtain a third image.

7. The method for detecting a surface defect of steel according to claim 5, wherein the obtaining a convolution weight for filtering according to the spatial proximity of the third image and the similarity of pixel values of the third image, and the filtering the third image according to the convolution weight to obtain a fourth image comprises:

defining a center point in said third image;

calculating the spatial proximity of each pixel point in the third image to the central point and the pixel value similarity of each pixel point and the central point;

multiplying the spatial proximity of each pixel point by the pixel value similarity to obtain a convolution weight of each pixel point;

and obtaining a fourth image according to the convolution weight value of each pixel point and the pixel value of each pixel point.

8. The steel surface defect detection method according to claim 5, wherein the acquiring the horizontal gradient image and the vertical gradient image of the fourth image for edge enhancement processing, and obtaining the first image from the horizontal gradient image and the vertical gradient image, comprises:

acquiring a horizontal matrix corresponding to the fourth image in the horizontal direction and a vertical matrix corresponding to the fourth image in the vertical direction;

Carrying out plane convolution calculation on the horizontal matrix and the fourth image to obtain the horizontal gradient image;

carrying out plane convolution calculation on the vertical matrix and the fourth image to obtain the vertical gradient image;

and carrying out bit-wise OR operation on the horizontal gradient image and the vertical gradient image to obtain the first image.

9. A steel surface defect detection apparatus, comprising:

the generation module is used for generating a binarization segmentation threshold value of the steel surface image to be detected according to the front Jing Huidu mean value and the background gray level mean value of the steel surface image to be detected;

the first processing module is used for sequentially carrying out gray level equalization processing, filtering processing and edge enhancement processing on the steel surface image to be detected to obtain a first image;

the second processing module is used for carrying out binarization processing on the first image by utilizing the binarization segmentation threshold value to obtain a second image;

the detection module is used for generating a defect detection result of the steel surface image to be detected according to the second image and a pre-trained defect detection model; the defect detection model is obtained by training based on a preset training set, the preset training set comprises a plurality of steel surface defect images and steel surface defect-free images, and the steel surface defect images are corresponding to defect types.

10. An electronic device comprising a memory for storing a computer program and a processor for calling and running the computer program stored in the memory, characterized in that the processor implements the steps of the method according to any of the preceding claims 1-8 when the computer program is executed.