CN113177921A - Magnetic shoe surface defect detection method based on neural network - Google Patents
Magnetic shoe surface defect detection method based on neural network Download PDFInfo
- Publication number
- CN113177921A CN113177921A CN202110480208.8A CN202110480208A CN113177921A CN 113177921 A CN113177921 A CN 113177921A CN 202110480208 A CN202110480208 A CN 202110480208A CN 113177921 A CN113177921 A CN 113177921A
- Authority
- CN
- China
- Prior art keywords
- neural network
- network model
- module
- magnetic shoe
- training
- 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; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Analysis (AREA)
- Investigating Or Analyzing Materials By The Use Of Magnetic Means (AREA)
Abstract
The invention discloses a magnetic shoe surface defect detection method based on a neural network, and provides a magnetic shoe surface defect detection method based on a neural network. An end-to-end CNN structure is provided, which is called as a Fusion Feature CNN (FFCNN), wherein the FFCNN comprises three modules, namely a feature extraction module, a feature fusion module and a decision module. The feature extraction module is used for extracting features from different images. The feature fusion module fuses the features extracted by the feature extraction module. And the decision module predicts the label through the fusion features. On this basis, an attention mechanism is introduced to focus attention on representative parts, suppressing less important information. Compared with the traditional manual detection, the detection system has the advantages that the efficiency and the precision are improved for the defect detection of the surface of the magnetic shoe, and the detection cost is reduced.
Description
Technical Field
The invention relates to a magnetic shoe surface defect detection method based on a neural network, in particular to a magnetic shoe surface defect detection method based on a neural network.
Background
The magnetic shoe is a tile-shaped permanent magnet made of magnetic materials and is a key component of a stator or a rotor with a constant magnetic field in a permanent magnet motor. The quality of which directly affects the performance and life cycle of the permanent magnet motor. This can have serious consequences if tiles with surface defects are used without detection. Therefore, it is very important to develop a set of efficient and effective magnetic shoe surface detection technology.
Most manufacturers adopt a manual detection method for technical reasons, and the manual detection method has the following problems: (1) the manual detection has low efficiency, low precision, large subjective factor and easy fatigue. (2) The human eye has difficulty in recognizing minute defects. The traditional neural network detection cannot directly acquire inherent combined features from a plurality of images and cannot accurately identify surface defects.
Disclosure of Invention
In view of the defects of the prior art, the invention aims to provide a magnetic tile surface defect detection method based on a neural network.
In order to achieve the purpose, the invention adopts the following technical scheme:
a magnetic shoe surface defect detection method based on a neural network comprises the following steps:
s1, collecting magnetic shoe surface images, and dividing the magnetic shoe surface images into a training set and an experimental set;
s2, training the neural network model by using a training set to obtain a neural network model I;
s3, adding a feature fusion module in the trained neural network model I to obtain a neural network model II;
s4, introducing an attention mechanism into the neural network model II to obtain a neural network III;
and S5, detecting the experiment set by using a neural network model III to obtain a detection result, and judging whether the surface of the magnetic shoe has defects or not according to the detection result.
Further, the S1 further includes: and (3) carrying out image collection on the surfaces of the M magnetic tile samples from N directions by using a rotary image collection system, wherein each magnetic tile sample obtains N images, and M and N are integers larger than 0.
Further, the S2 further includes:
s21, classifying the training set according to different acquisition directions, classifying the images of different magnetic shoe samples in the same acquisition direction into one stream, and classifying G streams, wherein G is an integer greater than 0, and G is equal to M;
s22, inputting each stream into an original neural network model, and pre-training the original neural network model to obtain a pre-training neural network model;
s23, taking the weight of the pre-trained neural network model as an initial value of the neural network model, and training the neural network model until convergence to obtain a neural network model I;
further, the step S23 further includes: using CBAM (convolutional Block attention Module) to suppress unwanted information, the loss is calculated using the following equation:
wherein m represents a minimum batch size; k represents a class number; l {. denotes an indication function, where l { true } ═ 1, and l { false } ═ 0.
Further, the S3 further includes: and fusing the characteristic fusion module with the neural network model I by using a connection operator.
Further, the join operator is a join operator.
Further, the attention mechanism is an attention module in a convolutional layer, the attention module in the convolutional layer comprising a channel attention module and a spatial attention module.
Further, the original neural network model is a convolutional neural network model, and the neural network model is a fusion feature convolutional neural network model.
The invention has the beneficial effects that:
the invention can replace a low-efficiency manual detection method, improve the recognition effect of the traditional neural network and achieve the aims of saving cost and improving production efficiency.
Drawings
FIG. 1 is a schematic diagram of the FFCNN architecture of the present invention
FIG. 2 is a schematic flow diagram of the present invention;
FIG. 3 is a schematic structural diagram of a CBAM according to the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings and the detailed description below:
in order to achieve the purpose, the invention adopts the following technical scheme:
the embodiment of the invention is as follows: a magnetic tile surface defect detection method based on an FFCNN model (Fused feature Convolutional Neural Network model) of a Resnet-50 (residual Network) architecture is disclosed, wherein the FFCNN model is an end-to-end CNN model and is called as a Fused feature CNN, and as shown in figure 1, the FFCNN consists of three modules, namely a feature extraction module, a feature fusion module and a decision module. The feature extraction module is used for extracting features from different images. The feature fusion module fuses the features extracted by the feature extraction module. And the decision module predicts the label through the fusion features. On this basis, an attention mechanism is introduced to focus attention on representative parts, suppressing less important information. Compared with the traditional manual detection, the detection system has the advantages that the efficiency and the precision are improved for the defect detection of the surface of the magnetic shoe, and the detection cost is reduced.
The embodiment comprises the following steps:
and S1, collecting the magnetic shoe surface image, and dividing the magnetic shoe surface image into a training set and an experimental set.
Step S1 further includes:
and (3) carrying out image collection on the surfaces of 1300 magnetic tile samples from N directions by using a rotary image acquisition system, wherein each magnetic tile sample obtains N images, and M and N are integers larger than 0.
Step S2, training the FFCNN model by using a training set to obtain an FFCNN model I;
as shown in fig. 1, the architecture of FFCNN includes a feature extraction module, a feature fusion module, and a decision module. The feature extraction module is composed of a plurality of convolution kernel convolution block attention modules. And the feature fusion module realizes feature fusion through feature mapping. The decision module includes multiple convolution kernels, multiple convolution attention modules, Avgpool (average pooling layer), and an FC2 layer.
Step S2 further includes:
s21, classifying the training set according to different acquisition directions, classifying the images of different magnetic shoe samples in the same acquisition direction into one stream, and classifying N streams;
s22, inputting each stream into a CNN model, and pre-training the CNN model to obtain a pre-trained CNN model;
s23, taking the weight of the pre-trained CNN model as an initial value of the neural network model, and training the FFCNN model until convergence to obtain the neural network model I.
Step S2 is directed to training the neural network.
According to the image labeling method, there were 1300 samples, of which there were 1000 defective samples and 300 normal samples. Each sample contained 4 images. The segmentation of the data set is shown in table 1. The initial learning rate is set to 0.001 and the mini-batch size is set to 10. Because the sample contains 4 images, FFCNN consists of 4 streams. As shown in algorithm 1, to train FFCNNs, a conventional Neural network (Convolutional Neural network) is first trained on each stream. And taking the weight of the pre-trained network as an initial value. After convergence of each stream, the learned weight value of each stream is used as an initial value of FFCNN. FFCNN is then jointly trained until convergence.
Table 1 distribution of experimental aluminium samples
| Sample set | Defect sample | Normal sample |
| Training set | 800 | 200 |
| Experimental suite | 200 | 100 |
As shown in algorithm 1, to train FFCNN, the training of conventional CNN is first performed separately for each stream. And taking the weight of the pre-trained network as an initial value. After convergence of each stream, the learned weight value of each stream is used as an initial value of FFCNN. FFCNN is then jointly trained until convergence.
FFCNN algorithm 1 is as follows:
input: training set C ═ X1，X2,……XK},
In the first step, each stream is trained to initially: θ is { θ 1, θ 2 … … θ n }, learning rate η, degree of element is Nt, and maximum batch is m. Learning rates were adjusted using Adam optimization algorithm.
For i＝1:n
The number of While elements is equal to or less than Nt, the ith stream is trained using m images, and the feature map is calculated using the following equation.
Γi＝{Si j|j＝1,2,...,N}＝Λ(Ii|θi)∈RW×H×N (1)
Where Γ denotes the learned feature map Sj denotes the jth feature of the image Ii; n represents a feature map of channel number learning of the image; Λ represents a feature extraction operation; θ i denotes a parameter of the ith stream; w, H and N represent the channel width, height and number of learned feature maps, respectively.
The loss is calculated using the following equation using CBAM (convolutional block attention module) to suppress the unwanted information.
Wherein m represents a minimum batch size; k represents a class number; l { · } denotes an indication function, where l { true } ═ 1, and no, l { false } ═ 0.
The parameter thetai values are updated using the SGD algorithm.
End while
Outputθi
End for
Outputθ＝{θ1,θ2,……θn}
Second, training FFCNN
Initial value: θ is { θ 1, θ 2 … … θ n } (using the output value of Step 1), learning rate η is Nt, the number of elements is Nt, and the maximum batch is m.
The number of While elements is equal to or less than Nt, FFCNN is trained using m sample maps, feature maps are calculated using equation (1) above, unnecessary information is suppressed using CBAM, and features are fused to a specified place using averaging, maximization, and join operators. The loss is calculated using equation (2) above, and the value of θ i is updated using the SGD algorithm.
End while
Outputθ＝{θ1，θ2……θm}。
Step S3, adding a feature fusion module in the trained FFCNN model I to obtain an FFCNN model II;
step S3 further includes:
and performing fusion of the feature fusion module and the FFCNNI by using a connection operator.
A feature fusion module is added to the last convolutional layer of the Resnet-50 framework, namely Cov5-X, feature fusion is carried out by using a containment operator, and experiments of fusion and combination of different layers are carried out. The feature fusion module is further introduced, a space fusion strategy is embedded in a traditional CNN architecture, and for a sample X, the feature extraction module learns features from each image Ii to obtain corresponding feature mapping xi. The feature fusion function may be represented by:
F＝θ(Г1，Г2，……Гn) (3)
wherein F is the output of the feature fusion module; θ () is a feature fusion function.
Three fusion operators were studied, including a mean operator, a maximum operator, and a join operator. Finally, a join fusion operator is selected, for which the learned feature maps are superimposed along the channel direction, which can be represented by the following formula.
Wherein Fcat is ∈ RwxHxN.
An architecture based on the Resnet-50 flow is shown in the following table:
step S4, introducing an attention mechanism into the FFCNNI to obtain FFCNNI;
step S4 further includes:
the Attention mechanism is CBAM (Attention Module in Convolutional layer), which includes a channel Attention Module and a space Attention Module.
Introducing an attention mechanism, CBAM is integrated into FFCNN, i.e. both channel attention module and space attention module are used, as shown in FIG. 3 below. (a) Is a channel attention module, and (b) is a space attention module. To further explain these two modules in conjunction with fig. 3, in the channel attention block, an indirect feature map is generated from the intermediate feature map Ft, and two pool operators, i.e., an average and a maximum operator, are applied to the special axis of the intermediate feature map Ft, and two vectors Vavg and Vmax are generated through the maximum pool operation and the average pool operation. The two vectors are then passed into a single-layer multi-layer perception (MLP), yielding attention vectors Aavg and Amax. And then combining the attention vector and the space attention vector according to the sum of the elements to extract the channel characteristics. And finally, inputting the output characteristic vector into a sigmoid activation function to obtain a channel attention vector Vc. Vc can be calculated by the following function:
in the formula (I), the compound is shown in the specification,
is a pool operator;
is the sum of intelligent elements; σ (-) denotes the sigmoid activation function.
As shown in fig. 3, in the spatial attention module, first, an average pool operation and a maximum pool operation are performed at a channel axis of a channel attention output feature map by channel feature extraction, and an attention matrix Mavg Mmax is generated. Then, it will be noted that the matrices Mavg and Mmax are subjected to a join operation to generate a feature matrix. To emphasize or suppress where the code is, a convolutional layer application performs convolutional operations, and finally performs spatial attention matrix operations. In short, the spatial attention matrix MS can be calculated by:
in the formula (I), the compound is shown in the specification,
is a feature map of the attention channel output; f. of7X7Is a convolution operation with a filter size of 7X 7.
And S5, detecting the experimental set by using the FFCNN model III to obtain a detection result, and judging whether the surface of the magnetic shoe has defects or not according to the detection result.
Various other modifications and changes may be made by those skilled in the art based on the above-described technical solutions and concepts, and all such modifications and changes should fall within the scope of the claims of the present invention.
Claims (8)
1. A magnetic tile surface defect detection method based on a neural network is characterized by comprising the following steps:
s1, collecting magnetic shoe surface images, and dividing the magnetic shoe surface images into a training set and an experimental set;
s2, training the neural network model by using a training set to obtain a neural network model I;
s3, adding a feature fusion module in the trained neural network model I to obtain a neural network model II;
s4, introducing an attention mechanism into the neural network model II to obtain a neural network III;
and S5, detecting the experiment set by using a neural network model III to obtain a detection result, and judging whether the surface of the magnetic shoe has defects or not according to the detection result.
2. The method for detecting the surface defect of the magnetic tile based on the neural network as claimed in claim 1, wherein the step S1 further comprises: and (3) carrying out image collection on the surfaces of the M magnetic tile samples from N directions by using a rotary image collection system, wherein each magnetic tile sample obtains N images, and M and N are integers larger than 0.
3. The method for detecting the surface defect of the magnetic tile based on the neural network as claimed in claim 1, wherein the step S2 further comprises:
s21, classifying the training set according to different acquisition directions, classifying the images of different magnetic shoe samples in the same acquisition direction into one stream, and classifying G streams, wherein G is an integer greater than 0, and G is equal to M;
s22, inputting each stream into an original neural network model, and pre-training the original neural network model to obtain a pre-training neural network model;
s23, taking the weight of the pre-trained neural network model as an initial value of the neural network model, and training the neural network model until convergence to obtain the neural network model I.
4. The method for detecting the surface defect of the magnetic tile based on the neural network as claimed in claim 3, wherein the step S23 further comprises: using CBAM (convolutional Block attention Module) to suppress unwanted information, the loss is calculated using the following equation:
wherein m represents a minimum batch size; k represents a class number; l {. denotes an indication function, where l { true } ═ 1, and l { false } ═ 0.
5. The method for detecting the surface defect of the magnetic tile based on the neural network as claimed in claim 1, wherein the step S3 further comprises: and fusing the characteristic fusion module with the neural network model I by using a connection operator.
6. The method as claimed in claim 5, wherein the join operator is a join operator.
7. The method of claim 1, wherein the attention mechanism is an attention module in a convolutional layer, and the attention module in the convolutional layer comprises a channel attention module and a space attention module.
8. The method as claimed in claim 1, wherein the original neural network model is a convolutional neural network model, and the neural network model is a fusion feature convolutional neural network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110480208.8A CN113177921A (en) | 2021-04-30 | 2021-04-30 | Magnetic shoe surface defect detection method based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110480208.8A CN113177921A (en) | 2021-04-30 | 2021-04-30 | Magnetic shoe surface defect detection method based on neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113177921A true CN113177921A (en) | 2021-07-27 |
Family
ID=76925790
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110480208.8A Pending CN113177921A (en) | 2021-04-30 | 2021-04-30 | Magnetic shoe surface defect detection method based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113177921A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109829891A (en) * | 2019-01-02 | 2019-05-31 | 浙江大学 | A kind of magnetic tile surface defect detection method based on intensive generation confrontation neural network |
| CN110619638A (en) * | 2019-08-22 | 2019-12-27 | 浙江科技学院 | Multi-mode fusion significance detection method based on convolution block attention module |
| CN110827263A (en) * | 2019-11-06 | 2020-02-21 | 创新奇智(南京)科技有限公司 | Magnetic shoe surface defect detection system and detection method based on visual identification technology |
| WO2020113561A1 (en) * | 2018-12-07 | 2020-06-11 | 华为技术有限公司 | Method for extracting structural data from image, apparatus and device |
| CN112164048A (en) * | 2020-09-25 | 2021-01-01 | 华中科技大学 | Magnetic shoe surface defect automatic detection method and device based on deep learning |
| CN112330594A (en) * | 2020-10-13 | 2021-02-05 | 浙江华睿科技有限公司 | Paper tube defect detection method and device, electronic equipment and storage medium |
-
2021
- 2021-04-30 CN CN202110480208.8A patent/CN113177921A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020113561A1 (en) * | 2018-12-07 | 2020-06-11 | 华为技术有限公司 | Method for extracting structural data from image, apparatus and device |
| CN109829891A (en) * | 2019-01-02 | 2019-05-31 | 浙江大学 | A kind of magnetic tile surface defect detection method based on intensive generation confrontation neural network |
| CN110619638A (en) * | 2019-08-22 | 2019-12-27 | 浙江科技学院 | Multi-mode fusion significance detection method based on convolution block attention module |
| CN110827263A (en) * | 2019-11-06 | 2020-02-21 | 创新奇智(南京)科技有限公司 | Magnetic shoe surface defect detection system and detection method based on visual identification technology |
| CN112164048A (en) * | 2020-09-25 | 2021-01-01 | 华中科技大学 | Magnetic shoe surface defect automatic detection method and device based on deep learning |
| CN112330594A (en) * | 2020-10-13 | 2021-02-05 | 浙江华睿科技有限公司 | Paper tube defect detection method and device, electronic equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| LUOFENG XIE: ""FFCNN: A Deep Neural Network for Surface Defect Detection of Magnetic Tile"", 《IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 》 * |
| 李婷: ""基于深度卷积神经网络的水工隧洞表面缺陷检测"", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107133943B (en) | A kind of visible detection method of stockbridge damper defects detection | |
| CN110827251B (en) | Power transmission line locking pin defect detection method based on aerial image | |
| Mayr et al. | Weakly supervised segmentation of cracks on solar cells using normalized L p norm | |
| CN111639564B (en) | Video pedestrian re-identification method based on multi-attention heterogeneous network | |
| CN101169867B (en) | Image dividing method, image processing apparatus and system | |
| CN114972312A (en) | Improved insulator defect detection method based on YOLOv4-Tiny | |
| CN112233074A (en) | Power failure detection method based on visible light and infrared fusion image | |
| CN115861190A (en) | Comparison learning-based unsupervised defect detection method for photovoltaic module | |
| CN114140672A (en) | Target detection network system and method applied to multi-sensor data fusion in rainy and snowy weather scene | |
| CN114581486A (en) | Template updating target tracking algorithm based on full convolution twin network multilayer characteristics | |
| CN104537381A (en) | Blurred image identification method based on blurred invariant feature | |
| CN116452862A (en) | Image classification method based on domain generalization learning | |
| CN117237884A (en) | Interactive inspection robot based on berth positioning | |
| Tang et al. | Deep Learning-Based Algorithm for Multi-Type Defects Detection in Solar Cells with Aerial EL Images for Photovoltaic Plants. | |
| CN111199255A (en) | Small target detection network model and detection method based on dark net53 network | |
| CN114359167A (en) | Insulator defect detection method based on lightweight YOLOv4 in complex scene | |
| CN117475137A (en) | Compact small target detection method based on composite feature map | |
| CN113177921A (en) | Magnetic shoe surface defect detection method based on neural network | |
| CN117036429A (en) | Infrared-visible light heterologous image registration method for photovoltaic defect detection | |
| CN117115616A (en) | Real-time low-illumination image target detection method based on convolutional neural network | |
| CN116543432A (en) | Efficient face detection method based on auxiliary network | |
| CN117132802A (en) | Method, device and storage medium for identifying field wheat diseases and insect pests | |
| CN113657310A (en) | Photovoltaic module infrared hot spot fault identification method based on multilayer convolutional neural network | |
| CN114581407A (en) | Self-adaptive defect detection method for photovoltaic module | |
| CN114299354A (en) | Insulator detection positioning method based on rotating frame identification network |
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 |