CN113569835A - Water meter numerical value reading method based on target detection and segmentation identification - Google Patents
Water meter numerical value reading method based on target detection and segmentation identification Download PDFInfo
- Publication number
- CN113569835A CN113569835A CN202110935454.8A CN202110935454A CN113569835A CN 113569835 A CN113569835 A CN 113569835A CN 202110935454 A CN202110935454 A CN 202110935454A CN 113569835 A CN113569835 A CN 113569835A
- Authority
- CN
- China
- Prior art keywords
- image
- segmentation
- detection
- target
- water meter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a water meter value reading method based on target detection and segmentation identification, and belongs to the field of computer vision deep learning. The method comprises a target image segmentation detection model consisting of a numerical value region target detection module, a target region segmentation module and a numerical value identification reading module, wherein the numerical value region target detection module is used for carrying out real-time detection to obtain the position information of a numerical value region in a water meter image; the target area segmentation module intercepts the water meter image according to the numerical value position information to obtain a numerical value area image, and performs image segmentation processing on the area of the water meter image after equally dividing the area by using a deep Lab model to obtain an image only with numbers; and finally, reading the correct reading of the water meter by using the SVM model. The invention effectively solves the problems of low reading identification efficiency, low accuracy and poor robustness.
Description
Technical Field
The invention relates to the fields of computer vision, machine learning and the like, in particular to a water meter numerical value reading method based on target detection and segmentation identification.
Background
In recent years, with the further development of image processing technology, many students have proposed an intelligent meter reading technology with low cost and high efficiency aiming at the defects of expensive labor cost and time management cost of manual meter reading. The artificial intelligence identification technology of the computer replaces the traditional method for reading the content, so that the production and labor efficiency of workers can be greatly improved, but the accuracy of the reading of the water meter is closely related to the specific environment of the water meter, the environmental conditions are different, and the accuracy of the identification is difficult to guarantee.
Although there has been much research on water meter reading identification, there are some unsolved problems. At present, most of researches on digital identification adopt a shallow machine learning model technology to perform pattern identification, for example, a digital character identification method based on a voting strategy, which is proposed by linyang et al, uses Hough transformation to locate the area where the water meter is located, then uses affine transformation technology to locate the position where the water meter reading is located, and finally uses the voting strategy to perform classification identification on the reading. Because shallow network information is less, once a large amount of dirt is attached to the water meter dial or a large amount of water drops exist in the water meter dial, the extraction of the water meter image characteristics is greatly influenced, and the characteristic point matching calculation amount between the water meter image and the template is large, so that the time is consumed very much. In the process of identification, because the dial plate of the water meter can reflect light, ambient light is also a big factor influencing the identification effect.
Therefore, in order to solve the problems, the invention provides a water meter value reading method based on target detection and segmentation identification. The method has certain advantages in the accuracy rate of numerical value reading and time cost.
Disclosure of Invention
In order to solve the above problems, the present invention provides a water meter value reading method based on target detection and segmentation identification. The efficiency and the accuracy can be improved by utilizing the target image segmentation detection model to carry out numerical region positioning and numerical content reading. The system is composed of a numerical value area target detection module, a target area segmentation module and a numerical value identification reading module.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a water meter value reading method based on target detection and segmentation identification comprises the following steps:
step 1: acquiring a water meter image in an actual application scene as a sample data set, and labeling a digital region and a numerical value pixel point in the sample water meter image;
step 2: establishing a target image segmentation detection model, wherein the target image segmentation detection model consists of a numerical value region target detection module, a target region segmentation module and a numerical value identification reading module;
and step 3: training a target image segmentation detection model by using the sample data set acquired in the step 1, specifically:
step 3.1: taking the marked water meter image as the input of a target image segmentation detection model, firstly acquiring a detection frame of a digital reading area in the water meter image through a numerical value area target detection module, then acquiring the detection frame of the digital reading area through a target area segmentation module, then intercepting an interested area from an original water meter image to obtain an image of a reading area and carrying out equal-proportion segmentation on the image to obtain an image only with a number, and finally generating and outputting the identified number through a numerical value identification reading module;
step 3.2: respectively training to obtain target detection loss of a numerical value region target detection module, segmentation image detection loss of a target region segmentation module and numerical value identification reading loss of a numerical value identification reading module by taking a numerical region and numerical value pixel points labeled in a sample water meter image as labels; taking the target detection loss, the segmentation image detection loss and the numerical value identification reading loss as total loss to finish the training of the target image segmentation detection model;
and 4, step 4: and acquiring a water meter image in real time, and taking the real-time image as the input of the trained target image segmentation detection model to obtain the numerical reading in the image.
Furthermore, the target region segmentation module comprises an equal proportion segmentation module, a ResNet residual error neural network, a cavity convolution network, a connecting layer, a convolution layer and an upper sampling layer; according to the coordinate value of the detection frame of the digital reading area, intercepting the region of interest from the original water meter image to obtain an image of the reading area, and cutting the image of the reading area into a fixed number of single digital images in an equal proportion along the long edge by an equal proportion cutting module; aiming at all single digital images, shallow layer features of the single digital images are extracted by using a ResNet residual neural network, high layer features of the single digital images are extracted by using a hole convolution network, the shallow layer features and the high layer features of the same single digital image are connected, and then the image only containing single numbers is obtained through a convolution layer and an upper sampling layer.
Compared with the traditional shallow machine learning model technology, the water meter reading identification method has the following advantages:
compared with the prior art, the target detection module adopts an Faster R-CNN network model to position the numerical region, thereby reducing the interference of light, dust, water drops and other environmental factors, improving the characteristic extraction capability and reducing the time cost; after positioning, obtaining a numerical image by utilizing an interested area for extracting the numerical value of the original water meter image, and reducing the interference of a large amount of useless information in the background; the target region segmentation module adopts a DeepLab V3 network model to extract a numerical image, and the advantage of a receptive field is increased under the condition that information is not lost through cavity convolution, so that the numerical image can be accurately extracted; and performing multi-classification operation on the numerical value image through an SVM support vector machine model, and finally obtaining the correct numerical value reading.
Drawings
FIG. 1 is a diagram of a target image segmentation detection model network architecture;
FIG. 2 is an overall flow diagram of the present method;
FIG. 3 is a diagram of steps in an embodiment of the method;
FIG. 4 is a flow chart of the fast R-CNN real-time detection of a numerical target in a water meter image;
FIG. 5 is a DeepLab network segmenting a numerical image;
FIG. 6 is a data set sample.
Detailed Description
The invention is described in detail below with reference to the drawings and specific embodiments, but the invention is not limited thereto.
A water meter value reading method based on target detection and segmentation identification, as shown in fig. 1-2, comprising the following steps:
step 1: acquiring a water meter image in an actual application scene as a sample data set, and labeling a digital region and a numerical value pixel point in the sample water meter image;
step 2: establishing a target image segmentation detection model, wherein the target image segmentation detection model consists of a numerical value region target detection module, a target region segmentation module and a numerical value identification reading module;
and step 3: training a target image segmentation detection model by using the sample data set acquired in the step 1, specifically:
step 3.1: taking the marked water meter image as the input of a target image segmentation detection model, firstly acquiring a detection frame of a digital reading area in the water meter image through a numerical value area target detection module, then acquiring the detection frame of the digital reading area through a target area segmentation module, then intercepting an interested area from an original water meter image to obtain an image of a reading area and carrying out equal-proportion segmentation on the image to obtain an image only with a number, and finally obtaining a corresponding reading through a numerical value identification reading module;
the numerical target detection module comprises a feature extraction layer, an RPN network layer, a feature fusion layer and a feature prediction layer;
firstly, obtaining a feature map from the water meter image through a feature extraction layer integrating convolution, an activation function and pooling; generating a corresponding candidate region through an RPN network layer, fusing the feature map and the candidate region through a feature fusion layer, and judging positive and negative samples by anchors; and finally, obtaining a series of detection frames of each feature map, and coordinate values and confidence degrees of each detection frame through a feature prediction layer, screening all the detection frames, and taking the coordinate values of the screened detection frames as the final output of the numerical target detection module. The screening of all detection frames specifically comprises the following steps:
1) pre-filtering: filtering out detection boxes with confidence scores lower than a threshold;
2) removing weight: carrying out maximum suppression treatment on the pre-filtered detection frames, and removing repeated detection frames;
3) and extracting the detection frame with higher target occurrence probability and the coordinate value thereof.
And the target area segmentation module intercepts an interested area from the original water meter image according to the detection frame coordinate value of the numerical area to obtain a numerical image.
Step 3.2: taking a numerical value region and numerical value pixel points marked in a sample water meter image as labels, and respectively obtaining the target detection loss of a numerical value region target detection module, the segmentation image detection loss of a target region segmentation module and the numerical value identification reading loss of a numerical value identification reading module; taking the target detection loss, the segmentation image detection loss and the numerical value identification reading loss as total loss to finish the training of the target image segmentation detection model;
and 4, step 4: and acquiring a water meter image in real time, and taking the real-time image as the input of the trained target image segmentation detection model to obtain the numerical reading in the image.
In the invention, a numerical value region target detection module adopts a Faster R-CNN model, a target region segmentation module adopts a deep Lab V3 model, and a numerical value reading detection module adopts an SVM support vector machine, as shown in FIG. 3, the implementation steps are as follows:
A. the fast R-CNN locates the numerical target in the water meter image:
firstly, initializing the Faster R-CNN, reading a parameter file, analyzing a Faster R-CNN model, and loading model weight.
As shown in FIG. 4, the fast R-CNN real-time detection target process: step A, synchronizing video image data after image scaling to a GPU, entering a backbone network of Faster R-CNN, and performing alternating processing of 13 convolutions, 13 activation functions and 4 pooling to obtain corresponding characteristic graphs. And then, sending the feature map into an RPN to generate a candidate region, judging whether anchors belong to positive samples or negative samples through a softmax function, and correcting the anchors by using frame regression to obtain an accurate candidate region. And then inputting the feature map and the candidate region into the region-of-interest pooling layer together to obtain comprehensive information, judging the feature position, and performing frame regression again to obtain the feature map for final prediction.
Each anchor of the candidate region will generate 9 prediction boxes, and each prediction box has 5 parameters, which respectively represent the coordinates of the upper left corner, the lower right corner, the width, the height and the confidence of the prediction box. And filtering the prediction frame with low confidence score by setting a threshold, finally carrying out maximum suppression on the reserved prediction frame to remove repeated frames, selecting the prediction frame with higher target occurrence probability, and outputting specific coordinates of the prediction frame.
B. Intercepting the original water meter image according to the coordinates of the numerical value area target to obtain a numerical value image:
and D, intercepting an interested region of a numerical value according to the specific coordinate obtained in the step A in the current water meter image to obtain a numerical value image, and carrying out geometric segmentation on the numerical value image.
C. The target area segmentation module performs semantic segmentation on the numerical value interesting area to obtain an image only with numerical values:
as shown in fig. 5, the numerical image obtained in step B is used to extract the shallow feature of the numerical value by using the ResNet residual neural network, and then the high-level feature is extracted by using the ASPP structure, and then the obtained shallow feature and the high-level feature are combined together, and then the feature map to be subjected to semantic segmentation finally, that is, the fixed number of images only containing a single number, is obtained by convolution, activation function and upsampling. .
D. And (3) reading the content of the numerical image according to a numerical identification reading module to obtain the final water meter reading:
and C, traversing each pixel point in the images according to the fixed number of images only containing a single number obtained in the step C, performing multi-classification operation through a supervised learning algorithm of an SVM (support vector machine), obtaining a final prediction result of each image only containing a single number, and splicing all the results in sequence to obtain a specific reading in the water meter images.
The effects of the present invention will be described below with reference to specific examples.
Data set: to evaluate the performance of the algorithm herein, a total of 1000 pictures were taken as a data set (as shown in fig. 6), and the data set was set as 4: the ratio of 1 randomly takes 800 pictures as a training set and 200 pictures as a test set. After the model training is finished, 200 images are tested.
Experimental parameters:
setting an initial learning rate to be 0.0001, adopting a strategy of adjusting the learning rate as required, setting the total iteration number to be 40000, adjusting the learning rate when the iteration number is 28000 and 32000, setting the decline rate gamma to be 0.1, setting batch to be 64, updating the network weight in an SGD mode, and adopting L2 regularization to avoid model overfitting, wherein momentum is set to be 0.9, and decay is set to be 0.0005. The following steps are training target image segmentation detection model network steps.
And (4) judging the standard: and evaluating the detection effect of the target detection and segmentation recognition image detection model by using Recall (Recall), Accuracy (Accuracy) and Frame Per Second (Frame Per Second FPS) indexes.
Experimental data: the recall rate is 82.55 percent, the accuracy rate is 96.8 percent, and 47fps is obtained. Therefore, the pointer reading recognition efficiency is high, and the accuracy is high.
The foregoing lists merely illustrate specific embodiments of the invention. It is obvious that the invention is not limited to the above embodiments, but that many variations are possible. All modifications which can be derived or suggested by a person skilled in the art from the disclosure of the present invention are to be considered within the scope of the invention.
Claims (7)
1. A water meter value reading method based on target detection and segmentation identification is characterized by comprising the following steps:
step 1: acquiring a water meter image in an actual application scene as a sample data set, and labeling a digital region and a numerical value pixel point in the sample water meter image;
step 2: establishing a target image segmentation detection model, wherein the target image segmentation detection model consists of a numerical value region target detection module, a target region segmentation module and a numerical value identification reading module;
and step 3: training a target image segmentation detection model by using the sample data set acquired in the step 1, specifically:
step 3.1: taking the marked water meter image as the input of a target image segmentation detection model, firstly acquiring a detection frame of a digital reading area in the water meter image through a numerical value area target detection module, then acquiring the detection frame of the digital reading area through a target area segmentation module, then intercepting an interested area from an original water meter image to obtain an image of a reading area and carrying out equal-proportion segmentation on the image to obtain an image only with a number, and finally generating and outputting the identified number through a numerical value identification reading module;
step 3.2: respectively training to obtain target detection loss of a numerical value region target detection module, segmentation image detection loss of a target region segmentation module and numerical value identification reading loss of a numerical value identification reading module by taking a numerical region and numerical value pixel points labeled in a sample water meter image as labels; taking the target detection loss, the segmentation image detection loss and the numerical value identification reading loss as total loss to finish the training of the target image segmentation detection model;
and 4, step 4: and acquiring a water meter image in real time, and taking the real-time image as the input of the trained target image segmentation detection model to obtain the numerical reading in the image.
2. The water meter value reading method based on the target detection and the segmentation identification as claimed in claim 1, wherein the value region target detection module comprises a feature extraction layer, an RPN network layer, a feature fusion layer and a feature prediction layer;
firstly, obtaining a feature map from the water meter image through a feature extraction layer integrating convolution, an activation function and pooling; generating a corresponding candidate region through an RPN network layer, fusing the feature map and the candidate region through a feature fusion layer, and judging positive and negative samples; and finally, obtaining a series of detection frames of each feature map, and coordinate values and confidence degrees of each detection frame through a feature prediction layer, screening all the detection frames, and taking the coordinate values of the screened detection frames as the final output of the numerical value region target detection module.
3. The water meter value reading method based on target detection and segmentation identification as claimed in claim 1, wherein the screening is performed on all detection frames, specifically:
1) pre-filtering: filtering out detection boxes with confidence scores lower than a threshold;
2) removing weight: carrying out maximum suppression treatment on the pre-filtered detection frames, and removing repeated detection frames;
3) and extracting the detection frame with higher target occurrence probability and the coordinate value thereof.
4. The water meter value reading method based on target detection and segmentation identification as claimed in claim 1, wherein the value region target detection module employs fast R-CNN model.
5. The water meter value reading method based on target detection and segmentation identification as claimed in claim 1, wherein the target area segmentation module comprises an equal proportion segmentation module, a ResNet residual neural network, a cavity convolution network, a connection layer, a convolution layer and an upper sampling layer; according to the coordinate value of the detection frame of the digital reading area, intercepting the region of interest from the original water meter image to obtain an image of the reading area, and cutting the image of the reading area into a fixed number of single digital images in an equal proportion along the long edge by an equal proportion cutting module; aiming at all single digital images, shallow layer features of the single digital images are extracted by using a ResNet residual neural network, high layer features of the single digital images are extracted by using a hole convolution network, the shallow layer features and the high layer features of the same single digital image are connected, and then the image only containing single numbers is obtained through a convolution layer and an upper sampling layer.
6. The method for reading the value of the water meter based on the target detection and the segmentation identification as claimed in claim 1, wherein the target area segmentation module adopts a deep model.
7. The water meter numerical value reading method based on the target detection and the segmentation recognition as claimed in claim 5, wherein the numerical value recognition reading module adopts an SVM (support vector machine) model, a fixed number of images only containing a single number are sequentially used as the input of the SVM model, a multi-classification prediction method is used to obtain the final prediction result of each image only containing a single number, and all the results are spliced in sequence to be used as the final recognized water meter numerical value result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110935454.8A CN113569835A (en) | 2021-08-16 | 2021-08-16 | Water meter numerical value reading method based on target detection and segmentation identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110935454.8A CN113569835A (en) | 2021-08-16 | 2021-08-16 | Water meter numerical value reading method based on target detection and segmentation identification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113569835A true CN113569835A (en) | 2021-10-29 |
Family
ID=78171696
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110935454.8A Pending CN113569835A (en) | 2021-08-16 | 2021-08-16 | Water meter numerical value reading method based on target detection and segmentation identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113569835A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114267044A (en) * | 2021-12-24 | 2022-04-01 | 三川智慧科技股份有限公司 | Data identification method and device for digital water meter |
| CN114359239A (en) * | 2022-01-07 | 2022-04-15 | 大连华信计算机技术股份有限公司 | Method and device for identifying dial reading and storage medium |
| CN114743188A (en) * | 2022-04-06 | 2022-07-12 | 上海电气集团股份有限公司 | Digital instrument identification method, system, electronic device and storage medium |
| CN115294574A (en) * | 2022-07-26 | 2022-11-04 | 山东浪潮科学研究院有限公司 | Remote water meter value reading method and system |
| CN118072312A (en) * | 2024-04-19 | 2024-05-24 | 深圳拓安信物联股份有限公司 | Water meter reading method, device, equipment and storage medium |
| CN120014622A (en) * | 2025-04-21 | 2025-05-16 | 杭州数聚链科技有限公司 | Ship water gauge card identification method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107729896A (en) * | 2017-10-24 | 2018-02-23 | 武汉科技大学 | A kind of automatic identifying method suitable for the wide tuning drive gear formula instrument of transformer station |
| CN109543688A (en) * | 2018-11-14 | 2019-03-29 | 北京邮电大学 | A kind of novel meter reading detection and knowledge method for distinguishing based on multilayer convolutional neural networks |
| WO2020181685A1 (en) * | 2019-03-12 | 2020-09-17 | 南京邮电大学 | Vehicle-mounted video target detection method based on deep learning |
| CN113191352A (en) * | 2021-03-23 | 2021-07-30 | 浙江理工大学 | Water meter pointer reading identification method based on target detection and binary image detection |
-
2021
- 2021-08-16 CN CN202110935454.8A patent/CN113569835A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107729896A (en) * | 2017-10-24 | 2018-02-23 | 武汉科技大学 | A kind of automatic identifying method suitable for the wide tuning drive gear formula instrument of transformer station |
| CN109543688A (en) * | 2018-11-14 | 2019-03-29 | 北京邮电大学 | A kind of novel meter reading detection and knowledge method for distinguishing based on multilayer convolutional neural networks |
| WO2020181685A1 (en) * | 2019-03-12 | 2020-09-17 | 南京邮电大学 | Vehicle-mounted video target detection method based on deep learning |
| CN113191352A (en) * | 2021-03-23 | 2021-07-30 | 浙江理工大学 | Water meter pointer reading identification method based on target detection and binary image detection |
Non-Patent Citations (1)
| Title |
|---|
| 尹远: "水表数字识别的研究与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》, pages 138 - 1592 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114267044A (en) * | 2021-12-24 | 2022-04-01 | 三川智慧科技股份有限公司 | Data identification method and device for digital water meter |
| CN114359239A (en) * | 2022-01-07 | 2022-04-15 | 大连华信计算机技术股份有限公司 | Method and device for identifying dial reading and storage medium |
| CN114743188A (en) * | 2022-04-06 | 2022-07-12 | 上海电气集团股份有限公司 | Digital instrument identification method, system, electronic device and storage medium |
| CN115294574A (en) * | 2022-07-26 | 2022-11-04 | 山东浪潮科学研究院有限公司 | Remote water meter value reading method and system |
| CN118072312A (en) * | 2024-04-19 | 2024-05-24 | 深圳拓安信物联股份有限公司 | Water meter reading method, device, equipment and storage medium |
| CN120014622A (en) * | 2025-04-21 | 2025-05-16 | 杭州数聚链科技有限公司 | Ship water gauge card identification method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114663346B (en) | A strip steel surface defect detection method based on improved YOLOv5 network | |
| CN113569835A (en) | Water meter numerical value reading method based on target detection and segmentation identification | |
| CN106875381B (en) | Mobile phone shell defect detection method based on deep learning | |
| CN114419014B (en) | Surface defect detection method based on feature reconstruction | |
| CN111460936A (en) | Remote sensing image building extraction method, system and electronic equipment based on U-Net network | |
| CN110929593B (en) | Real-time significance pedestrian detection method based on detail discrimination | |
| CN111428550A (en) | Vehicle detection method based on improved YO L Ov3 | |
| CN111598861A (en) | Detection method of non-uniform texture small defects based on improved Faster R-CNN model | |
| CN111950453A (en) | Optional-shape text recognition method based on selective attention mechanism | |
| CN110728694B (en) | Long-time visual target tracking method based on continuous learning | |
| CN112287941B (en) | License plate recognition method based on automatic character region perception | |
| CN114998603B (en) | An underwater target detection method based on deep multi-scale feature factor fusion | |
| CN110046617A (en) | A kind of digital electric meter reading self-adaptive identification method based on deep learning | |
| CN111652225A (en) | Non-invasive camera reading method and system based on deep learning | |
| Zhao et al. | Research on detection method for the leakage of underwater pipeline by YOLOv3 | |
| CN113191352A (en) | Water meter pointer reading identification method based on target detection and binary image detection | |
| CN112329771B (en) | Deep learning-based building material sample identification method | |
| CN109902662A (en) | A pedestrian re-identification method, system, device and storage medium | |
| CN117437647B (en) | Oracle bone text detection method based on deep learning and computer vision | |
| CN113487610A (en) | Herpes image recognition method and device, computer equipment and storage medium | |
| CN116416503A (en) | Small sample target detection method, system and medium based on multi-mode fusion | |
| CN118506263A (en) | A foreign body detection method for power transmission lines in complex environments based on deep learning | |
| CN110458064B (en) | Combining data-driven and knowledge-driven low-altitude target detection and recognition methods | |
| CN114937239B (en) | Pedestrian multi-target tracking identification method and tracking identification device | |
| CN110991374A (en) | A fingerprint singularity detection method based on RCNN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20211029 |
|
| RJ01 | Rejection of invention patent application after publication |