CN114022420A - Detection method for automatically identifying defects of photovoltaic cell EL assembly - Google Patents
Detection method for automatically identifying defects of photovoltaic cell EL assembly Download PDFInfo
- Publication number
- CN114022420A CN114022420A CN202111228565.1A CN202111228565A CN114022420A CN 114022420 A CN114022420 A CN 114022420A CN 202111228565 A CN202111228565 A CN 202111228565A CN 114022420 A CN114022420 A CN 114022420A
- Authority
- CN
- China
- Prior art keywords
- photovoltaic cell
- network model
- defect
- component
- image
- 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.)
- Granted
Links
- 230000007547 defect Effects 0.000 title claims abstract description 106
- 238000001514 detection method Methods 0.000 title claims abstract description 64
- 230000004927 fusion Effects 0.000 claims abstract description 29
- 238000000605 extraction Methods 0.000 claims abstract description 12
- 238000012545 processing Methods 0.000 claims abstract description 12
- 238000012549 training Methods 0.000 claims description 36
- 238000012360 testing method Methods 0.000 claims description 23
- 238000007781 pre-processing Methods 0.000 claims description 16
- 238000010606 normalization Methods 0.000 claims description 13
- 238000010586 diagram Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 9
- 238000002372 labelling Methods 0.000 claims description 9
- 230000009467 reduction Effects 0.000 claims description 7
- 238000011176 pooling Methods 0.000 claims description 5
- 238000013519 translation Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000002360 preparation method Methods 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 2
- 230000007246 mechanism Effects 0.000 abstract description 7
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000000034 method Methods 0.000 abstract description 6
- 238000003062 neural network model Methods 0.000 abstract description 5
- 230000008569 process Effects 0.000 abstract description 3
- 238000013135 deep learning Methods 0.000 abstract description 2
- 230000005764 inhibitory process Effects 0.000 abstract 1
- 238000005401 electroluminescence Methods 0.000 description 65
- 230000002950 deficient Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000005520 cutting process Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 239000000969 carrier Substances 0.000 description 2
- 238000007670 refining Methods 0.000 description 2
- 229910000679 solder Inorganic materials 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- 206010027146 Melanoderma Diseases 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000000712 assembly Effects 0.000 description 1
- 238000000429 assembly Methods 0.000 description 1
- 208000003464 asthenopia Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 229910021419 crystalline silicon Inorganic materials 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 229910021421 monocrystalline silicon Inorganic materials 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
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
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- 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)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Signal Processing (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention discloses a detection method for automatically identifying defects of a photovoltaic cell EL component, which is based on an improved YOLO v5s neural network model, adds a self-designed residual error channel attention gating mechanism module suitable for detecting the defects of the photovoltaic cell EL component in a feature fusion part in an original model, guides multi-scale feature fusion and highlights complex background features in a defect position area inhibition image, and effectively improves the identification capability of the defects of the photovoltaic cell EL component. The detection method combines the deep learning technology and the image processing technology, thereby not only avoiding the low efficiency and uncertainty of the traditional manual feature extraction, but also having stronger robustness in the detection process; when the initial learning rate is set to be 0.001, the average accuracy of model classification reaches 94.1%, the detection precision is obviously improved, and the detection speed is improved.
Description
Technical Field
The invention belongs to the technical field of defect detection, and particularly relates to a detection method for automatically identifying defects of a photovoltaic cell EL assembly.
Background
The global energy crisis is a very concerned problem for governments of all countries, the traditional fossil energy is not only non-renewable, but also brings great pollution to the earth, and the development of novel renewable energy is a global subject which lasts for decades.
Solar energy is a large pillar of new energy, and can be converted into heat energy, chemical energy and electric energy, wherein the theoretical basis for converting solar energy into electric energy is the photovoltaic effect, which is discovered by french scientist beckerel in 1839. In 1954, scientists in Bell laboratories in the United states produced practical single crystal silicon solar cells for the first time. Through the development of nearly 70 years, solar cells have been applied to various fields from the top of aviation and the bottom of daily life, the production technology of the solar cells is more and more mature, and the photovoltaic solar industry is rapidly developed.
The photovoltaic cell generates various defects in the production process, the defects influence the photoelectric conversion rate of the cell and determine the grading of the cell pieces during sale, so that the detection of the cell defects is necessary. Electroluminescence (EL) is an important technology of battery imaging, and the principle is that a forward bias voltage is applied to a crystalline silicon solar battery, a large number of unbalanced carriers are injected into the battery, the unbalanced carriers are continuously compounded to emit light to form infrared rays, then the infrared rays are captured by an industrial camera, and the infrared rays are processed by a computer and displayed into an image.
Various defects are generated during the production of photovoltaic EL modules. According to the size of the defect, the battery can be divided into different grades to meet the requirements of different scenes, most of the process still depends on manual judgment, and the following defects exist: (1) the mastery degree of different quality inspectors on the standard is not uniform. (2) The image defect positions of the EL components are random and have more types, and quality inspectors cannot fully identify the defects, so that the defects are often missed. (3) The quality inspector can easily generate visual fatigue and make wrong or missed judgment when watching a large number of pictures for a long time. (4) One qualified quality inspector needs to be trained systematically and professionally, and the culture period is long.
Therefore, it becomes a work of great research for professionals to effectively improve and solve the difficulties caused by the manual evaluation method. Under the condition, a detection method capable of intelligently detecting the defects of the photovoltaic cell EL assembly is developed, so that the overall quality of the photovoltaic cell EL assembly is improved, the overall cost of the photovoltaic industry is reduced, and the overall development of the photovoltaic industry is promoted.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a detection method for automatically identifying the defects of a photovoltaic cell EL component. The detection method is an automatic defect identification detection method of the photovoltaic cell EL component improved based on a YOLOv5s neural network model, and has high defect identification accuracy and high detection speed.
The technical scheme adopted by the invention for solving the technical problems is as follows: a detection method for automatically identifying defects of an EL component of a photovoltaic cell is designed, and is characterized by comprising the following steps:
the first step is as follows: image pre-processing
1) Acquiring a defect image library of the photovoltaic cell EL assembly: firstly, acquiring an original image of a photovoltaic cell EL component by an industrial camera, and then carrying out operations such as rotation, translation, contrast deepening and the like on the original image to obtain a preprocessed image; finally, carrying out size normalization operation on the original image and the preprocessed image, eliminating images without defects, and forming a defect image library of the photovoltaic cell EL assembly by the rest images;
2) sample set data preparation: manually and randomly selecting 50-80% of defect image libraries of the photovoltaic cell EL assembly as training sample sets, wherein the rest are test sample sets, respectively labeling images in each sample set, and adding defect type labels;
the second step is that: training of photovoltaic cell EL component defect detection network model based on improved Yolov5s network model
1) Sample set preprocessing
Preprocessing a training sample set in a Mosaic data enhancement mode;
2) parameter setting
Initializing all weight values, bias values and batch normalization scale factor values, setting the initial learning rate and batch _ size of the network, inputting initialized parameter data into the network, and setting the initial learning rate of the network to be 0.001;
3) network model training
Inputting the preprocessed training sample set into an improved Yolov5s network model with set initialization parameters, wherein the network model is obtained by replacing a Concat module in the Yolov5s network model with an RCAG module; under the processing of an improved Yolov5s network model, firstly extracting features and combining with an RCAG module to perform multi-scale fusion, then refining and fusing feature maps obtained by extracting the last three layers by a convolutional layer to reduce dimensionality, and predicting tensors by using a classification regression network, wherein the predicted values comprise positions, confidence degrees and classification scores; comparing the generated predicted value with the label information to generate a loss value, then performing reverse propagation, updating parameters of a backbone network and a classification regression network until the loss value is in accordance with the preset value, and finishing the training of the network model parameters;
4) network model testing
Inputting the test sample set into the network model which completes parameter training in the step 3) to obtain a tensor prediction value of the test sample set; comparing the predicted value of the tensor with the labeling information, testing the reliability of the network model, and monitoring whether the model is over-fitted so as to determine whether the training needs to be stopped and parameters are readjusted;
the third step: photovoltaic cell EL assembly defect detection
And (3) subjecting the photovoltaic cell EL component image to be detected to the same size normalization operation in the step 1) in the first step, and then inputting the photovoltaic cell EL component image to be detected to be reliable in the step two, so as to obtain defect tensor information of the photovoltaic cell EL component image to be detected, wherein the defect tensor information comprises a defect position, a defect type and a confidence coefficient.
Compared with the prior art, the invention has the beneficial effects that:
the detection method is based on an improved YOLO v5s neural network model, a self-designed residual channel attention gating mechanism module suitable for detecting the defects of the photovoltaic cell EL component is added to a feature fusion part in an original model, multi-scale feature fusion is guided, complex background features in a defect position area suppression image are highlighted, and the recognition capability of the defects of the photovoltaic cell EL component is effectively improved. The detection method combines the deep learning technology and the image processing technology, so that the low efficiency and uncertainty of the traditional manual feature extraction are avoided, and meanwhile, the detection process has strong robustness; the defects of the photovoltaic cell EL component are detected by adopting the original YOLOv5s network model, the average recognition rate is 90.2%, and when the initial learning rate is set to be 0.001 by adopting the detection method, the average accuracy rate of model classification reaches 94.1%, so that the detection precision is obviously improved, and the detection speed is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without any creative effort.
FIG. 1 is a flow chart of an embodiment of the detection method of the present invention.
Fig. 2 is a diagram of a network model structure for detecting defects of a photovoltaic cell EL assembly based on an improved Yolov5s network model in an embodiment of the detection method of the present invention.
Fig. 3 is a working schematic diagram of a Focus module of a photovoltaic cell EL component defect detection network model based on an improved Yolov5s network model according to an embodiment of the detection method of the present invention.
Fig. 4 is a schematic view of Mosaic enhancement of a photovoltaic cell EL component defect detection network model based on an improved Yolov5s network model in an embodiment of the detection method of the present invention.
Fig. 5 is a schematic structural diagram of a CSP1_ x module of a photovoltaic cell EL assembly defect detection network model based on an improved Yolov5s network model according to an embodiment of the detection method of the present invention.
Fig. 6 is a schematic structural diagram of a CSP2_ x module of a photovoltaic cell EL assembly defect detection network model based on an improved Yolov5s network model according to an embodiment of the detection method of the present invention.
Fig. 7 is a schematic structural diagram of a residual channel attention gating mechanism module of a photovoltaic cell EL component defect detection network model based on an improved Yolov5s network model according to an embodiment of the detection method of the present invention.
Fig. 8 is an image of a photovoltaic cell EL assembly to be tested.
FIG. 9 is a graph showing the results of the detection of the image of FIG. 8 using the detection method of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention relates to a detection method for automatically identifying defects of a photovoltaic cell EL assembly, which is used for detecting the photovoltaic cell EL assembly and comprises the following steps:
the first step is as follows: image pre-processing
1) Acquiring a defect image library of the photovoltaic cell EL assembly: firstly, acquiring an original image (including a defective image and a non-defective image) of a photovoltaic cell EL assembly by an industrial camera, and then performing operations such as rotation, translation, contrast deepening and the like on the original image to obtain a preprocessed image; finally, carrying out size normalization operation on the original image and the preprocessed image, eliminating images without defects, and forming a defect image library of the photovoltaic cell EL assembly by the rest images;
2) sample set data preparation: manually and randomly selecting 50-80% of defect image libraries of the photovoltaic cell EL assembly as training sample sets, wherein the rest are test sample sets, respectively labeling images in each sample set, and adding defect type labels;
secondly, training a photovoltaic cell EL component defect detection network model based on the improved Yolov5s network model
3) Sample set preprocessing
And preprocessing the training sample set in a Mosaic data enhancement mode.
The Mosaic data enhancement adopts the modes of random zooming, random cutting and random arrangement to splice the images, and has good detection effect on small targets. And then, adjusting the resolution of the original input image, and adaptively scaling the picture according to the different aspect ratio of the picture.
4) Parameter setting
Initializing all weight values, bias values and batch normalization scale factor values, setting the initial learning rate and batch _ size (batch processing parameters) of the network, inputting the initialized parameter data into the network, and setting the initial learning rate of the network to be 0.001;
3) network model training
Inputting the preprocessed training sample set into an improved Yolov5s network model with set initialization parameters, wherein the network model is obtained by replacing a Concat module (splicing module) in the Yolov5s network model with an RCAG module (residual channel attention gating mechanism module); under the processing of an improved Yolov5s network model, firstly, feature extraction is carried out on a preprocessed training sample set, multi-scale fusion is carried out by combining an RCAG module, then, feature maps obtained by the last three layers of extraction are subjected to thinning fusion by a convolution layer to reduce dimensionality, and tensors including position, confidence and classification score predicted values are predicted by utilizing a classification regression network. And comparing the generated predicted value with the label information to generate a loss value, then performing reverse propagation, and updating parameters of the backbone network and the classification regression network until the loss value is in accordance with the preset value, and finishing the training of the network model parameters.
4) Network model testing
Inputting the test sample set into the network model which completes parameter training in the step 3) to obtain a tensor prediction value of the test sample set; and comparing the tensor predicted value with the labeling information, and testing the reliability of the network model for monitoring whether the model is over-fitted or not so as to determine whether the training needs to be stopped and parameters need to be readjusted or not. When the accuracy rate of the tensor prediction value is more than 90%, the network model can be considered to be reliable.
The third step: photovoltaic cell EL assembly defect detection
And (3) subjecting the photovoltaic cell EL component image to be detected to the same size normalization operation in the step 1) in the first step, and then inputting the photovoltaic cell EL component image to be detected to be reliable in the step two, so as to obtain defect tensor information of the photovoltaic cell EL component image to be detected, wherein the defect tensor information comprises a defect position, a defect type and a confidence coefficient.
Example 1
The embodiment provides a detection method for automatically identifying and detecting defects of a photovoltaic cell EL assembly, which is used for detecting the photovoltaic cell EL assembly and comprises the following steps:
the first step is as follows: image pre-processing
1) Acquiring a defect image library of the photovoltaic cell EL assembly: firstly, acquiring an original image (including a defective image and a non-defective image) of a photovoltaic cell EL assembly by an industrial camera, and then performing operations such as rotation, translation, contrast deepening and the like on the original image to obtain a preprocessed image; finally, carrying out size normalization operation on the original image and the preprocessed image, eliminating images without defects, and forming a defect image library of the photovoltaic cell EL assembly by the rest images;
because the number of the defect images collected by normal shooting is obviously less than that of the normal images, the collected photovoltaic cell EL component images are subjected to preprocessing operations such as rotation, translation, contrast deepening and the like to expand a photovoltaic cell EL component database; then, carrying out size normalization operation on all the collected photovoltaic cell EL component images and the photovoltaic cell EL component images obtained by preprocessing to obtain images with uniform size, and finally removing unnecessary parts (without defects) in the images by adopting regional morphological processing to form a photovoltaic cell EL component defect image library;
the original picture size used in this embodiment is 3600 × 618, and the picture size sent to training is 641 × 589; since the size of the original picture is 3600 × 618 and the pixel value is too large, each picture is cut into 6 pieces, the size of the cut picture is 641 × 589, the images without defects are removed, and finally 7440 pieces of defect image libraries of the photovoltaic cell EL assemblies are obtained.
2) Sample set data preparation: manually and randomly selecting 80% of the defect image library of the photovoltaic cell EL assembly as a training sample set, using the rest 20% of the defect image library as a test sample set, respectively labeling images in each sample set, and adding a defect type label;
the marked object is an image defect area (including black spots, cold solder joints, hidden cracks, broken grids and linear defects), and the defect area is marked manually by using LabelImg; in this embodiment, the training sample set and the test sample set select 7440 defective pictures, wherein 1488 black spots are marked as heiban; 1488 cold joints marked as xuhan; 1488 subfractures marked yinlie; a broken grid 1488 is marked as Duanshan; line defects 1488, labeled xianzhuangquexian.
Secondly, training a photovoltaic cell EL component defect detection network model based on the improved Yolov5s network model
1) Sample set preprocessing
And preprocessing the training sample set in a Mosaic data enhancement mode.
The Mosaic data enhancement adopts the modes of random zooming, random cutting and random arrangement to splice the images, and has good detection effect on small targets. And then, adjusting the resolution of the original input image, and adaptively scaling the picture according to the different aspect ratio of the picture.
2) Parameter setting
Initializing all weight values, bias values and batch normalization scale factor values, setting the initial learning rate of the network and the batch _ size to be 64, inputting initialized parameter data into the network, and setting the initial learning rate of the network to be 0.001;
the model parameters are initialized as follows: the maximum number of iterations (epoch) is set to 200, the first 50 epoch learning rates are set to 0.001, the last 150 epoch learning rates are set to 0.0001, the descent factor of the learning rate is 0.1, and the weight decay of the regularization term is 0.0005.
3) Network model training
Inputting the preprocessed training sample set into an improved Yolov5s network model with set initialization parameters, wherein the network model is obtained by replacing a Concat module (splicing module) in the Yolov5s network model with an RCAG module (residual channel attention gating mechanism module); under the processing of an improved Yolov5s network model, firstly, feature extraction is carried out on a preprocessed training sample set, multi-scale fusion is carried out by combining an RCAG module, then, feature maps obtained by the last three layers of extraction are subjected to thinning fusion by a convolution layer to reduce dimensionality, and a classified regression network is used for predicting tensors, including position, defect types and confidence coefficient prediction values. And comparing the generated predicted value with the labeling information to generate a loss value, then performing reverse propagation, and updating parameters of the backbone network and the classification regression network until the loss value is in accordance with the preset value, and finishing the training of the network model parameters.
4) Network model testing
Preprocessing a test sample set in a Mosaic data enhancement mode in the same step 1), and inputting the preprocessed test sample set into a network model which completes parameter training in the step 3) to obtain a tensor predicted value of the test sample set; and comparing the tensor prediction value with the labeling information to obtain the accuracy rate of the single type defects which is more than 90 percent, and showing in table 1 in detail, which proves that the network model is an effective model.
TABLE 1 test results
Kind of defect | Black spot | Insufficient solder joint | Hidden crack | Broken grid | Line defect |
Rate of accuracy | 96.2% | 94.3% | 91.6% | 93.4% | 91.0% |
The third step: photovoltaic cell EL assembly defect detection
The method comprises the steps of carrying out the same size normalization operation on a photovoltaic cell EL component image to be detected in the step 1) of the first step, preprocessing the photovoltaic cell EL component image to be detected in a Mosaic data enhancement mode in the step 1) of the second step, and inputting the preprocessed image into a photovoltaic cell EL component defect detection network model tested to be reliable in the step two to obtain defect tensor information of the photovoltaic cell EL component image to be detected, wherein the defect tensor information comprises a defect position, a defect type and a confidence coefficient.
The defect detection method adopts an improved Yolov5s neural network model, has high detection precision and high identification speed: specifically, the neural network model structure is shown in fig. 3, and mainly comprises a data enhancement part, a feature extraction part, a feature fusion part and a classification regression part. The data enhancement part adopts the Mosaic data enhancement, splices the images of the photovoltaic cell EL assembly in the modes of random zooming, random cutting and random arrangement, and improves the detection capability of small targets. In the aspect of feature extraction, images of photovoltaic cell EL components are input into a Focus structure and a CSPDarknet network, and a feature map is extracted from the images through operations such as convolution, pooling and the like, wherein the feature map can be shared for a subsequent feature fusion part. The feature fusion part adopts FPN (feature pyramid) and PAN (pyramid attention network structure) to generate pooling feature vectors with different fixed sizes, so that the feature expression capability is enhanced, and the method has good effect on detection of the same object in different sizes. The classification regression part uses GIOU _ Loss as a Loss function of the Bounding box, effectively solves the problem of non-coincidence of the boundary frames, and improves the speed and the precision of the regression of the prediction frame.
The characteristic extraction stage of YOLOv5s adopts Focus and CSPDarknet53 structures, the original 640 x3 image is input into the Focus structure, and is changed into 320 x 12 characteristic diagram by adopting the slicing operation, and finally into 320 x 32 characteristic diagram by 32 convolution operation of convolution kernel. The CSPDarknet53 is formed by adding a CSP module (cross-phase local module) consisting of a convolutional layer and a residual structure Concate in a Resnet network to the darkey 53. The darkey 53 totals 53 layers of convolution, removing the last FC (full connection layer, actually realized by 1x1 convolution) for a total of 52 convolutions to serve as the host network. The 52 convolutional layers are composed of: the convolution kernel of 1 32 filters is firstly carried out, and then 5 repeated residual unit resblock _ body (the 5 residual units, each unit is composed of 1 independent convolution layer and a group of repeatedly executed convolution layers, the repeatedly executed convolution layers are respectively repeated for 1 time, 2 times, 8 times and 4 times, in each repeatedly executed convolution layer, the convolution operation of 1x1 is firstly executed, then the convolution operation of 3x3 is executed, the number of the filters is firstly halved and then restored), and the total number is 52.
The CSP Peaknet uses CSP modules on the basis of Darknet53, two structures of CSP1_ x and CSP2_ x are designed in Yolov5s, wherein the feature extraction part comprises CSP1_ x modules, and the feature fusion part comprises CSP2_ x modules. The convolution kernel size in front of each CSP module is 3 × 3, the step size is 2, and the function of down-sampling can be achieved. A CSP1_1 structure and a CSP1_3 structure are adopted in a backbone network, a CSP2_1 structure is adopted in a neck, the input of the CSP _ x is divided into two parts, one part is firstly subjected to x times of residual error operation and then convolution operation, the other part is directly subjected to convolution operation, the convolution operation and the residual error operation are both used for reducing the number of channels by half, then the two parts are spliced and output to strengthen the characteristic fusion between networks, the learning capacity of CNN is enhanced, and the accuracy is kept while the weight is reduced.
Design of RCAG module (residual channel attention gating mechanism module): aiming at the feature fusion part of low-level features and high-level features in the original network structure, the low-level features contain rich position information, the high-level features contain rich semantic information, but channels are treated equally in the fusion, so that the representation capability of the network is hindered. The RCAG module designed by the invention adds the up-sampled low-level features and the high-level features, strengthens the fusion of the features through convolution operation, predicts the importance of each channel through global average pooling operation and a full connection layer to obtain the importance of different channels, and can improve the sensitivity of the model to the channel features by multiplying the weight values to the strengthened features. And then, short connection in a residual error structure is added on the basis, so that richer feature information can be allowed to be directly propagated backwards through identity mapping, and the information flow can be ensured. By using a residual channel attention gating mechanism to replace a splicing operation in an FPN (feature pyramid) module and a PAN (path aggregation) module, multi-scale feature fusion can be guided, and complex background features in a defect position area suppression image can be highlighted.
The working principle of the RCAG module is as follows: the method comprises the steps of converting a low-level feature p after up-sampling and a high-level feature q into two feature spaces f and g through a convolution kernel of 1x1, obtaining a fusion feature S after element summation and ReLu function activation, and filtering the fusion feature S through convolution operation. Extracting global features and realizing dimension reduction work by using a Global Average Pooling (GAP) layer, inputting the global features into a multilayer perceptron (MLP) layer for processing, and then processing by a sigmoid function to generate a channel level attention diagram A; and multiplying the channel level attention diagram A by the fusion feature S to carry out feature re-weighting to obtain a weighted feature, and then carrying out pixel addition on the weighted feature and the fusion feature S through a residual connecting operation to obtain a final output feature. The MLP mainly comprises two full-connection layers (the first layer is provided with C/r channels, r is a reduction ratio and realizes dimension reduction operation, the second layer is provided with C channels) and a ReLu activation function, the two full-connection layers aim to model the relationship between the channels and can highlight an object and inhibit a complex background, the ReLu function is used for refining global features, the MLP is used for coding fusion features S, and the dependency between the channels is learned.
In the third step of this embodiment, a group of images of the photovoltaic cell EL assembly to be detected is shown in fig. 9, and the obtained detection structure is shown in fig. 9, where heiban, xuhan, yinlie, duanshan, xianzhuangquexian are all defect types, a frame represents a defect position, the defect position is accurately marked by a square frame, a number on the frame represents a confidence of a detection result, the maximum value is 1, and a higher numerical value represents a more accurate detection result.
The test result of the invention shows that the improved Yolov5s model is applied to the defect detection of the photovoltaic cell EL component, the single type accuracy rate of the defect detection of the photovoltaic cell EL component exceeds 90 percent, the average accuracy rate can reach 94.1 percent, and the rapid intelligent detection of the photovoltaic cell EL component can be realized.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
Nothing in this specification is said to apply to the prior art.
Claims (4)
1. A detection method for automatically identifying defects of a photovoltaic cell EL component is characterized by comprising the following steps:
the first step is as follows: image pre-processing
1) Acquiring a defect image library of the photovoltaic cell EL assembly: firstly, acquiring an original image of a photovoltaic cell EL component by an industrial camera, and then performing operations such as rotation, translation, contrast deepening and the like on the original image to obtain a preprocessed image; finally, carrying out size normalization operation on the original image and the preprocessed image, eliminating images without defects, and forming a defect image library of the photovoltaic cell EL assembly by the rest images;
2) sample set data preparation: manually and randomly selecting 50-80% of defect image libraries of the photovoltaic cell EL assembly as training sample sets, wherein the rest are test sample sets, respectively labeling images in each sample set, and adding defect type labels;
the second step is that: training of photovoltaic cell EL component defect detection network model based on improved Yolov5s network model
1) Sample set preprocessing
Preprocessing a training sample set in a Mosaic data enhancement mode;
2) parameter setting
Initializing all weight values, bias values and batch normalization scale factor values, setting the initial learning rate and batch _ size of the network, inputting initialized parameter data into the network, and setting the initial learning rate of the network to be 0.001;
3) network model training
Inputting the preprocessed training sample set into an improved Yolov5s network model with set initialization parameters, wherein the network model is obtained by replacing a Concat module in the Yolov5s network model with an RCAG module; under the processing of an improved Yolov5s network model, firstly, feature extraction is carried out on a preprocessed training sample set, multi-scale fusion is carried out by combining an RCAG module, then, feature maps obtained by the last three-layer extraction are subjected to thinning fusion by a convolution layer to reduce dimensionality, and a classified regression network is used for predicting tensors, wherein the predicted values comprise positions, confidence degrees and classified scores; comparing the generated predicted value with the label information to generate a loss value, then performing reverse propagation, updating parameters of a backbone network and a classification regression network until the loss value is in accordance with the preset value, and finishing the training of the network model parameters;
4) network model testing
Inputting the test sample set into the network model which completes parameter training in the step 3) to obtain a tensor prediction value of the test sample set; comparing the predicted value of the tensor with the labeling information, testing the reliability of the network model, and monitoring whether the model is over-fitted so as to determine whether to stop training and readjust parameters;
the third step: photovoltaic cell EL assembly defect detection
And (3) carrying out the same size normalization operation in the step 1) in the first step on the photovoltaic cell EL component image to be detected, and then inputting the photovoltaic cell EL component image to be detected into the photovoltaic cell EL component defect detection network model tested to be reliable in the step two, so as to obtain defect tensor information of the photovoltaic cell EL component image to be detected, wherein the defect tensor information comprises a defect position, a defect type and a confidence coefficient.
2. The detection method for automatically identifying the defects of the photovoltaic cell EL component as claimed in claim 1, wherein in the step 4) of the second step, the network model is considered to be reliable when the tensor prediction value accuracy rate is greater than 90%.
3. The detection method for automatically identifying the defects of the photovoltaic cell EL component as claimed in claim 1, wherein the RCAG module works according to the following principle: respectively converting the low-level features after up-sampling and the high-level features into two feature spaces through a 1x1 convolution kernel, obtaining fusion features after element summation and ReLu function activation, and then filtering the fusion features through convolution operation; then, extracting global features and realizing dimension reduction work by using a global average pooling layer, inputting the global features and the dimension reduction work into a multilayer perceptron layer for processing, and then processing by a sigmoid function to generate a channel-level attention diagram; and multiplying the channel-level attention diagram by the fusion feature to carry out feature re-weighting to obtain a weighted feature, and then carrying out pixel addition on the weighted feature and the fusion feature through a residual connecting operation to obtain a final output feature.
4. The detection method for automatically identifying the defects of the photovoltaic cell EL component as claimed in claim 3, wherein the multilayer perceptron layer mainly comprises two full-connected layers and a ReLu activation function, the first full-connected layer has C/r channels, r is a reduction ratio to realize the dimension reduction operation, and the second full-connected layer has C channels.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111228565.1A CN114022420B (en) | 2021-10-21 | 2021-10-21 | Detection method for automatically identifying defects of photovoltaic cell EL (electro-luminescence) component |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111228565.1A CN114022420B (en) | 2021-10-21 | 2021-10-21 | Detection method for automatically identifying defects of photovoltaic cell EL (electro-luminescence) component |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114022420A true CN114022420A (en) | 2022-02-08 |
CN114022420B CN114022420B (en) | 2024-05-31 |
Family
ID=80057050
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111228565.1A Active CN114022420B (en) | 2021-10-21 | 2021-10-21 | Detection method for automatically identifying defects of photovoltaic cell EL (electro-luminescence) component |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114022420B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114529546A (en) * | 2022-04-24 | 2022-05-24 | 科大天工智能装备技术(天津)有限公司 | Roof panel defect detection method and system |
CN114580568A (en) * | 2022-03-24 | 2022-06-03 | 华南理工大学 | Fish species identification method based on deep learning |
CN114594103A (en) * | 2022-04-12 | 2022-06-07 | 四川大学 | Method and system for automatically detecting surface defects of nuclear industrial equipment and automatically generating reports |
CN114820609A (en) * | 2022-06-29 | 2022-07-29 | 南昌大学 | Photovoltaic module EL image defect detection method |
CN115144406A (en) * | 2022-09-07 | 2022-10-04 | 广州珐玛珈智能设备股份有限公司 | Omnibearing online industrial visual detection and screening device and method |
CN115754108A (en) * | 2022-11-23 | 2023-03-07 | 福建省杭氟电子材料有限公司 | Acidity measuring system and method for electronic-grade hexafluorobutadiene |
CN116542980A (en) * | 2023-07-06 | 2023-08-04 | 宁德时代新能源科技股份有限公司 | Defect detection method, defect detection apparatus, defect detection program, storage medium, and defect detection program |
CN116912605A (en) * | 2023-09-12 | 2023-10-20 | 法施达(天津)智能科技有限公司 | Fault detection method and system based on anchor bolt image data feature recognition |
CN118037740A (en) * | 2024-04-15 | 2024-05-14 | 宁德时代新能源科技股份有限公司 | Pole piece detection method, device, equipment, storage medium and computer program product |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110570410A (en) * | 2019-09-05 | 2019-12-13 | 河北工业大学 | Detection method for automatically identifying and detecting weld defects |
CN112906794A (en) * | 2021-02-22 | 2021-06-04 | 珠海格力电器股份有限公司 | Target detection method, device, storage medium and terminal |
WO2021139069A1 (en) * | 2020-01-09 | 2021-07-15 | 南京信息工程大学 | General target detection method for adaptive attention guidance mechanism |
-
2021
- 2021-10-21 CN CN202111228565.1A patent/CN114022420B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110570410A (en) * | 2019-09-05 | 2019-12-13 | 河北工业大学 | Detection method for automatically identifying and detecting weld defects |
WO2021139069A1 (en) * | 2020-01-09 | 2021-07-15 | 南京信息工程大学 | General target detection method for adaptive attention guidance mechanism |
CN112906794A (en) * | 2021-02-22 | 2021-06-04 | 珠海格力电器股份有限公司 | Target detection method, device, storage medium and terminal |
Non-Patent Citations (1)
Title |
---|
周颖;毛立;张燕;陈海永;: "改进的CNN在太阳能电池板缺陷检测中的应用", 计算机仿真, no. 03, 15 March 2020 (2020-03-15) * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114580568A (en) * | 2022-03-24 | 2022-06-03 | 华南理工大学 | Fish species identification method based on deep learning |
CN114594103A (en) * | 2022-04-12 | 2022-06-07 | 四川大学 | Method and system for automatically detecting surface defects of nuclear industrial equipment and automatically generating reports |
CN114594103B (en) * | 2022-04-12 | 2023-05-16 | 四川大学 | Automatic detection and report generation method and system for surface defects of nuclear industrial equipment |
CN114529546A (en) * | 2022-04-24 | 2022-05-24 | 科大天工智能装备技术(天津)有限公司 | Roof panel defect detection method and system |
CN114820609A (en) * | 2022-06-29 | 2022-07-29 | 南昌大学 | Photovoltaic module EL image defect detection method |
CN115144406A (en) * | 2022-09-07 | 2022-10-04 | 广州珐玛珈智能设备股份有限公司 | Omnibearing online industrial visual detection and screening device and method |
CN115144406B (en) * | 2022-09-07 | 2022-11-11 | 广州珐玛珈智能设备股份有限公司 | Omnibearing online industrial visual detection and screening device and method |
CN115754108A (en) * | 2022-11-23 | 2023-03-07 | 福建省杭氟电子材料有限公司 | Acidity measuring system and method for electronic-grade hexafluorobutadiene |
CN116542980A (en) * | 2023-07-06 | 2023-08-04 | 宁德时代新能源科技股份有限公司 | Defect detection method, defect detection apparatus, defect detection program, storage medium, and defect detection program |
CN116542980B (en) * | 2023-07-06 | 2023-11-03 | 宁德时代新能源科技股份有限公司 | Defect detection method, defect detection apparatus, defect detection program, storage medium, and defect detection program |
CN116912605A (en) * | 2023-09-12 | 2023-10-20 | 法施达(天津)智能科技有限公司 | Fault detection method and system based on anchor bolt image data feature recognition |
CN118037740A (en) * | 2024-04-15 | 2024-05-14 | 宁德时代新能源科技股份有限公司 | Pole piece detection method, device, equipment, storage medium and computer program product |
Also Published As
Publication number | Publication date |
---|---|
CN114022420B (en) | 2024-05-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114022420B (en) | Detection method for automatically identifying defects of photovoltaic cell EL (electro-luminescence) component | |
CN111415329B (en) | Workpiece surface defect detection method based on deep learning | |
He et al. | Deep regression neural network for industrial surface defect detection | |
CN110610475B (en) | Visual defect detection method of deep convolutional neural network | |
CN110689011A (en) | Solar cell panel defect detection method of multi-scale combined convolution neural network | |
CN114973032B (en) | Deep convolutional neural network-based photovoltaic panel hot spot detection method and device | |
CN113837994B (en) | Photovoltaic panel defect diagnosis method based on edge detection convolutional neural network | |
Al-Mashhadani et al. | Deep learning methods for solar fault detection and classification: a review | |
CN115690675B (en) | ESB-YOLO model cultured fish swarm detection method based on channel non-dimension-reducing attention mechanism and improved YOLOv5 | |
Jiang et al. | Attention M-net for automatic pixel-level micro-crack detection of photovoltaic module cells in electroluminescence images | |
CN115908354A (en) | Photovoltaic panel defect detection method based on double-scale strategy and improved YOLOV5 network | |
Chindarkkar et al. | Deep learning based detection of cracks in electroluminescence images of fielded PV modules | |
Demirci et al. | Defective PV cell detection using deep transfer learning and EL imaging | |
Chen et al. | Automatic crack segmentation and feature extraction in electroluminescence images of solar modules | |
CN111612803B (en) | Vehicle image semantic segmentation method based on image definition | |
CN112179846A (en) | Prefabricated convex window defect detection system based on improved Faster R-CNN | |
Xu et al. | Defect detection for PV Modules based on the improved YOLOv5s | |
CN114612468B (en) | Equipment external defect detection method based on positive sample | |
CN116051530A (en) | Semi-supervised photovoltaic cell surface anomaly detection method based on image restoration | |
Singh et al. | Deep Segmenter system for recognition of micro cracks in solar cell | |
CN213957126U (en) | Prefabricated bay window defect detecting system | |
CN113780434B (en) | Deep learning-based solar cell module defect EL detection method | |
Tang et al. | Deep learning based model for Defect Detection of Mono-Crystalline-Si Solar PV Module Cells in Electroluminescence Images Using Data Augmentation | |
CN118379234A (en) | Photovoltaic module electroluminescent image defect identification method based on improvement Yolov s | |
CN117592518A (en) | YOLO-PV model for detecting defects of photovoltaic cell assembly and construction method thereof |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |