CN114021741A - Photovoltaic cell panel inspection method based on deep learning - Google Patents
Photovoltaic cell panel inspection method based on deep learning Download PDFInfo
- Publication number
- CN114021741A CN114021741A CN202111113670.0A CN202111113670A CN114021741A CN 114021741 A CN114021741 A CN 114021741A CN 202111113670 A CN202111113670 A CN 202111113670A CN 114021741 A CN114021741 A CN 114021741A
- Authority
- CN
- China
- Prior art keywords
- feature
- cell panel
- photovoltaic cell
- network
- weight
- 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
- 238000007689 inspection Methods 0.000 title claims abstract description 27
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000013135 deep learning Methods 0.000 title claims abstract description 19
- 230000006870 function Effects 0.000 claims abstract description 35
- 230000011218 segmentation Effects 0.000 claims abstract description 13
- 238000001514 detection method Methods 0.000 claims abstract description 11
- 238000012549 training Methods 0.000 claims abstract description 11
- 230000004927 fusion Effects 0.000 claims abstract description 10
- 238000000605 extraction Methods 0.000 claims abstract description 9
- 238000012360 testing method Methods 0.000 claims abstract description 9
- 238000003384 imaging method Methods 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims abstract description 6
- 230000004913 activation Effects 0.000 claims abstract description 5
- 238000005070 sampling Methods 0.000 claims description 30
- 238000010586 diagram Methods 0.000 claims description 25
- 230000007246 mechanism Effects 0.000 claims description 21
- 238000011176 pooling Methods 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000009826 distribution Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 230000002776 aggregation Effects 0.000 claims description 3
- 238000004220 aggregation Methods 0.000 claims description 3
- 230000000295 complement effect Effects 0.000 claims description 3
- 238000005336 cracking Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 2
- 230000007547 defect Effects 0.000 abstract description 6
- 238000005516 engineering process Methods 0.000 description 3
- 239000000463 material Substances 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 229910000838 Al alloy Inorganic materials 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000004566 building material Substances 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000002503 electroluminescence detection Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 239000005341 toughened glass Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Economics (AREA)
- Data Mining & Analysis (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a photovoltaic cell panel inspection method based on deep learning, which relates to the field of power inspection and image processing and comprises the following steps: preparing network input data, and acquiring an image data set of a photovoltaic cell panel by using an unmanned aerial vehicle carrying imaging equipment; the training set and the test set are enhanced and expanded through operations such as turning, rotating, mirroring, brightness adjusting and the like, and the diversity of data is increased; according to the invention, a detection algorithm of a defect area of the photovoltaic cell panel is designed through an improved Unet segmentation model, the network structure of an original Unet model is improved according to the characteristics of an image acquired by an unmanned aerial vehicle and the requirement of detecting a micro area, a double attention feature fusion module is introduced, and appropriate network parameters such as an activation function are set, so that the feature extraction of micro defects such as cracks is realized, the interference of a background is inhibited, the robustness and the accuracy of the network are improved, the problem of routing inspection of the photovoltaic cell panel can be effectively solved, and the stable operation of a photovoltaic transformer substation is guaranteed.
Description
Technical Field
The invention relates to the technical field of power inspection and image processing, in particular to a photovoltaic cell panel inspection method based on deep learning.
Background
With the further increase of energy crisis and environmental pollution, people are gradually seeking new green energy to replace the traditional fossil energy. Solar energy is used as a renewable energy source, and has the advantages of easy acquisition, sustainable utilization, environmental protection and the like. Meanwhile, the photovoltaic industry and technology are also continuously developed, and photovoltaic power stations are gradually built to utilize solar energy all over the world. In these power stations, a large number of solar panels are distributed, each of which is one of the most important components in solar power generation, and the conversion rate and the service life of the panels are important factors for determining whether the solar cells have use value. As a device for converting light energy into electric energy, the building materials of the solar cell panel comprise a solar cell panel, an EVA material, an aluminum alloy, a TP back panel, tempered glass and the like. Any unexpected error in the production process can cause defects, and meanwhile, the solar panel is exposed to severe environments such as exposure, rain and the like for a long time, and the defects inevitably cause damage to the panel in different degrees. How to timely find and deal with the problems directly influences the photoelectric conversion efficiency and the service life of the solar panel. The traditional detection method mainly adopts the inspection of artificial vision, an electroluminescence detection EL method, an ultrasonic detection method and the like. However, these detection methods require a lot of manpower, material resources and materials, and have very low efficiency, and cannot meet the actual requirements of industrial detection. With the rapid development of unmanned aerial vehicle technology, the unmanned aerial vehicle has been gradually applied to the fields of power inspection and the like. Carry on visible light or infrared integrative camera through industry unmanned aerial vehicle and patrol and examine large-scale photovoltaic power plant, can this a series of problems of effectual solution. The method comprises the steps of firstly, collecting a photovoltaic image to be detected by an unmanned aerial vehicle according to a planned routing inspection route, then, carrying out region segmentation by using technologies such as image processing and the like, segmenting a foreground region and a background region, and finally, carrying out fault detection and analysis to obtain a conclusion.
When the photovoltaic cell panel is patrolled and examined at present, a target photovoltaic cell panel region cannot be accurately and rapidly patrolled and examined, the true state of equipment cannot be obtained timely, and a fault region is found to complete an overhaul task, so that the stable operation of the whole photovoltaic transformer substation cannot be guaranteed.
Disclosure of Invention
The invention aims to solve the problems that when a photovoltaic cell panel is inspected, a target photovoltaic cell panel area cannot be accurately and quickly inspected, the real state of equipment cannot be obtained in time, and a fault area is found to finish an inspection task, so that the stable operation of the whole photovoltaic transformer substation cannot be guaranteed in the prior art, and provides a photovoltaic cell panel inspection method based on deep learning.
In order to achieve the purpose, the invention adopts the following technical scheme:
a photovoltaic cell panel inspection method based on deep learning comprises the following steps:
step 1: preparing network input data, and acquiring an image data set of a photovoltaic cell panel by using an unmanned aerial vehicle carrying imaging equipment;
step 2: the training set and the test set are enhanced and expanded through operations such as turning, rotating, mirroring, brightness adjusting and the like, and the diversity of data is increased;
and step 3: constructing a parallel attention mechanism optimized segmentation network U-Net;
and 4, step 4: constructing a parallel attention module;
and 5: training the data set constructed in the step 1 by using the constructed model;
step 6: and testing the images shot by the unmanned aerial vehicle by using the trained network, and finely adjusting the network.
Preferably, prepare network input data, utilize unmanned aerial vehicle to carry the image data set that imaging device gathered photovoltaic cell panel, including different damage types: the method comprises the steps of breaking, cracking, covering and the like, adjusting the size of all images to be 256 pixels multiplied by 256 pixels in eight bits, and then carrying out truth value labeling on the collected images through Labelme software and outputting labeled data sets and corresponding truth value graphs.
Preferably, the data set is enhanced through rotation, mirroring, scaling and brightness adjustment, the type and the number of the data set are enriched, and the enhanced data set is divided into a training set, a test set and a verification set according to the ratio of 6:3: 1.
Preferably, a parallel attention mechanism optimized segmentation network U-Net is constructed, the parallel attention mechanism optimized segmentation network U-Net mainly comprises an up-sampling part and a down-sampling part, the up-sampling part uses a VGG16 network to extract features, a first layer of network firstly adopts two groups of 3 x 3 convolutions, a batch normalization layer and a ReLu function layer to form a standard convolution module, a second layer adopts maximum pooling to perform down-sampling, the standard convolution is repeated, and a third layer to a fifth layer of network repeat three groups of convolution operations to extract deep semantic features;
when in down-sampling, the size of the feature map is reduced by half every time one layer is added, and the number of channels is multiplied by the number of channels of the fifth layer, wherein the number of the channels is not changed; in the up-sampling process, Dropout operation is carried out on the fifth-layer network with the probability of 0.5 to prevent overfitting of the network; the model adopts bilinear interpolation for up-sampling, and then the up-sampled model and the same-scale feature map from the fourth layer of the down-sampling stage are sent into a parallel attention mechanism module to enhance deep feature extraction; after the feature map of the ninth layer is obtained, 1 × 1 convolution is used in combination with Sigmoid operation, thereby outputting a segmentation result map of the network.
Preferably, a parallel attention module is constructed, and because a large amount of background interference exists in a photovoltaic cell panel detection task, a parallel double-attention mechanism is adopted to optimize a U-Net network for feature extraction; the parallel double attention module comprises the following three parts: channel attention module, spatial attention module, feature fusion.
Preferably, training is carried out on the data set constructed in the step 1 by using the constructed model; for input image data X, obtaining a characteristic diagram after encoding processingThe calculation formula is as follows:
in the formula: conv stands for convolution operation; function C represents the feature aggregation mechanism: convolution layer + normalization; the Down function represents Down-sampling; l represents the number of network layers; then decoding the obtained characteristic diagram to obtain the characteristic diagramThe calculation formula is as follows:
in the formula: [] Representing a concatenation of dimensions; up represents the Up-sampling of the function; bp represents a parallel attention mechanism operation; and (3) carrying out Sigmoid on the finally obtained feature graph to obtain a final prediction result:
the loss function is defined by a binary cross entropy loss function, defined as follows:
preferably, the channel attention module functions as follows:
firstly, defining the feature graph of an input module as F, and then obtaining a 1 × 1 × C global feature F after passing through a global maximum pooling layer1(ii) a Then using a one-dimensional convolution of size k on the global feature F1Performing convolution to obtain an attention weight map G; attention weight graph G weight mu of ith channeliCan be obtained by the following formula:
in the formula:j is more than or equal to 1 and less than or equal to k and represents the jth weight used for learning the ith channel weight;representing a set of global features of a k-th adjacent channel in the ith channel;
the convolution kernel size k of the one-dimensional convolution is calculated by:
in the formula: c is the number of channels; γ and b represent hyper-parameters, typically γ ═ 2, b ═ 1; | A |oddRepresents the nearest odd number adjacent to a; weighting the feature map F through the weight map G to obtain a feature map R; the weight of the channel related to the damage area to be detected is improved in the characteristic diagram R, so that the weight of other channels is reduced.
Preferably, the spatial attention module functions as follows:
and respectively performing average pooling and maximum pooling on the feature map F in a channel dimension to generate two feature maps of a single channel: characteristic graph of mean valueAnd maximum value feature mapThen, combining the feature graphs of the two single channels to generate a weight graph H and weighting the feature graph F to generate a feature graph S;
the spatial attention module calculates as follows:
wherein σ represents a ReLu function;representing the dot product of the corresponding pixels in both profiles.
Preferably, the feature fusion function is as follows:
adding the characteristic diagram R and the characteristic diagram S, and then obtaining a weight diagram G by using a ReLu activation function; the obtained weight graph is respectively fused with the weight distribution of the channel dimension and the weight distribution of the space dimension, so that more complementary features are obtained, and the feature regions such as damage, cracks and the like are concerned.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the invention, a detection algorithm of a defect area of the photovoltaic cell panel is designed through an improved Unet segmentation model, the network structure of an original Unet model is improved according to the characteristics of an image acquired by an unmanned aerial vehicle and the requirement of detecting a micro area, a double attention feature fusion module is introduced, and appropriate network parameters such as an activation function are set, so that the feature extraction of micro defects such as cracks is realized, the interference of a background is inhibited, the robustness and the accuracy of the network are improved, the problem of routing inspection of the photovoltaic cell panel can be effectively solved, and the stable operation of a photovoltaic transformer substation is guaranteed.
2. The invention introduces a parallel double-attention mechanism, designs a parallel double-attention feature fusion module in order to realize accurate distinguishing of micro areas under a complex background, and applies the parallel double-attention feature fusion module in the cross-layer connection with the same scale between the upper and lower samples, so that the network can pay more attention to the detection of micro damaged and cracked areas.
Drawings
Fig. 1 is an overall flow diagram of a photovoltaic cell panel inspection method based on deep learning according to the present invention;
fig. 2 is a schematic diagram of an overall network structure of a photovoltaic cell panel inspection method based on deep learning according to the present invention;
fig. 3 is a schematic network structure diagram of a dual-attention-machine system of the photovoltaic cell panel inspection method based on deep learning.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the 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.
Referring to fig. 1-3, a photovoltaic cell panel inspection method based on deep learning includes the following steps:
step 1: prepare network input data, utilize unmanned aerial vehicle to carry the image data set that imaging device gathered photovoltaic cell panel, including different damage types: breaking, cracking, covering and the like, adjusting the size of all images to be 256 pixels multiplied by 256 pixels in eight bits, and then carrying out truth value annotation on the acquired images through Labelme software and outputting an annotated data set and a corresponding truth value diagram;
step 2: enhancing the data set by rotating, mirroring, zooming and brightness adjusting, enriching the type and the number of the data set, and dividing the enhanced data set into a training set, a test set and a verification set according to the ratio of 6:3: 1;
and step 3: constructing a parallel attention mechanism optimized segmentation network U-Net; the segmented network U-Net network optimized by the parallel attention mechanism mainly comprises an up-sampling part and a down-sampling part, wherein the up-sampling part uses a VGG16 network to extract features, a layer 1 network firstly adopts two groups of 3 multiplied by 3 convolutions, a batch normalization layer and a ReLu function layer to form a standard convolution module, a second layer adopts maximum pooling to perform down-sampling, the standard convolution is repeated, and a third-fifth layer network repeats three groups of convolution operations to extract deep semantic features;
when in down-sampling, the size of the feature map is reduced by half every time one layer is added, and the number of channels is multiplied by the number of channels of the fifth layer, wherein the number of the channels is not changed; in the up-sampling process, Dropout operation is carried out on the fifth-layer network with the probability of 0.5 to prevent overfitting of the network; the model adopts bilinear interpolation for up-sampling, and then the up-sampled model and the same-scale feature map from the fourth layer of the down-sampling stage are sent into a parallel attention mechanism module to enhance deep feature extraction; after the feature map of the ninth layer is obtained, 1 × 1 convolution is used in combination with Sigmoid operation, so that a segmentation result map of the network is output;
and 4, step 4: constructing a parallel attention module; because a large amount of background interference exists in the detection task of the photovoltaic cell panel, the parallel double-attention mechanism is adopted to optimize the U-Net network for feature extraction; the parallel double attention module comprises the following three parts: a channel attention module, a spatial attention module and feature fusion;
the channel attention module functions as follows:
firstly, defining the feature graph of an input module as F, and then obtaining a 1 × 1 × C global feature F after passing through a global maximum pooling layer1(ii) a Then using a one-dimensional convolution of size k on the global feature F1Performing convolution to obtain a weight map G after attention reinforcement; attention weight graph G weight mu of ith channeliCan be obtained by the following formula:
in the formula:j is more than or equal to 1 and less than or equal to k and represents the jth weight used for learning the ith channel weight;representing a set of global features of a k-th adjacent channel in the ith channel;
the convolution kernel size k of the one-dimensional convolution is calculated by:
in the formula: c is the number of channels; γ and b represent hyper-parameters, typically γ ═ 2, b ═ 1; | A |oddRepresents the nearest odd number adjacent to a; weighting the feature map F through the weight map G to obtain a feature map R; the weight of the channel related to the damage area to be detected is improved in the characteristic diagram R, so that the weight of other channels is reduced.
The spatial attention module functions as follows:
and respectively performing average pooling and maximum pooling on the feature map F in a channel dimension to generate two feature maps of a single channel: characteristic graph of mean valueAnd maximum value feature mapThen, combining the feature graphs of the two single channels to generate a weight graph H and weighting the feature graph F to generate a feature graph S;
the spatial attention module calculates as follows:
wherein σ represents a ReLu function;representing the dot product of the corresponding pixels in both profiles.
The feature fusion function is as follows:
adding the characteristic diagram R and the characteristic diagram S, and then using a ReLu activation function to obtain a weight diagram G; the finally obtained weight graph is respectively fused with the weight distribution of the channel dimension and the weight distribution of the space dimension, so that more complementary characteristics are obtained, interference caused by various backgrounds can be inhibited, and characteristic regions such as damage and cracks can be concerned.
And 5: training the data set constructed in the step 1 by using the constructed model; for input image data X, obtaining a characteristic diagram after encoding processingThe calculation formula is as follows:
in the formula: conv stands for convolution operation; function C represents the feature aggregation mechanism: convolution layer + normalization; the Down function represents Down-sampling; l represents the number of network layers; then decoding the obtained characteristic diagram to obtain the characteristic diagramThe calculation formula is as follows:
in the formula: [] Representing a concatenation of dimensions; up represents the Up-sampling of the function; bp represents a parallel attention mechanism operation; and (3) carrying out Sigmoid on the finally obtained feature graph to obtain a final prediction result:
in the aspect of loss function, the invention selects a binary cross entropy loss function, which is defined as follows:
step 6: and testing the images shot by the unmanned aerial vehicle by using the trained network, and finely adjusting the network.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.
Claims (9)
1. The photovoltaic cell panel inspection method based on deep learning is characterized by comprising the following steps:
step 1: preparing network input data, and acquiring an image data set of a photovoltaic cell panel by using an unmanned aerial vehicle carrying imaging equipment;
step 2: the training set and the test set are enhanced and expanded through operations such as turning, rotating, mirroring, brightness adjusting and the like, and the diversity of data is increased;
and step 3: constructing a parallel attention mechanism optimized segmentation network U-Net;
and 4, step 4: constructing a parallel attention module;
and 5: training the data set constructed in the step 1 by using the constructed model;
step 6: and testing the images shot by the unmanned aerial vehicle by using the trained network, and finely adjusting the network.
2. The photovoltaic cell panel inspection method based on deep learning of claim 1, wherein network input data are prepared, and an unmanned aerial vehicle carrying imaging equipment is used for acquiring an image data set of the photovoltaic cell panel, wherein the image data set comprises different damage types: the method comprises the steps of breaking, cracking, covering and the like, adjusting the size of all images to be 256 pixels multiplied by 256 pixels in eight bits, and then carrying out truth value labeling on the collected images through Labelme software and outputting labeled data sets and corresponding truth value graphs.
3. The photovoltaic cell panel inspection method based on deep learning of claim 1 is characterized in that data set enhancement is performed through rotation, mirroring, scaling and brightness adjustment, the types and the number of the data sets are enriched, and the enhanced data sets are divided into a training set, a test set and a verification set according to the ratio of 6:3: 1.
4. The photovoltaic cell panel inspection method based on deep learning of claim 1 is characterized in that a parallel attention mechanism optimized segmentation network U-Net is constructed, the parallel attention mechanism optimized segmentation network U-Net mainly comprises an up-sampling part and a down-sampling part, the up-sampling part uses a VGG16 network to perform feature extraction, a first layer of network firstly adopts two groups of 3 x 3 convolutions, a batch normalization layer and a ReLu function layer to form a standard convolution module, a second layer adopts maximum pooling to perform down-sampling, the standard convolution is repeated, and third to fifth layers of networks repeat three groups of convolution operations to extract deep semantic features;
when in down-sampling, the size of the feature map is reduced by half every time one layer is added, and the number of channels is multiplied by the number of channels of the fifth layer, wherein the number of the channels is not changed; in the up-sampling process, Dropout operation is carried out on the fifth-layer network with the probability of 0.5 to prevent overfitting of the network; the model adopts bilinear interpolation for up-sampling, and then the up-sampled model and the same-scale feature map from the fourth layer of the down-sampling stage are sent into a parallel attention mechanism module to enhance deep feature extraction; after the feature map of the ninth layer is obtained, 1 × 1 convolution is used in combination with Sigmoid operation, thereby outputting a segmentation result map of the network.
5. The photovoltaic cell panel inspection method based on deep learning of claim 1 is characterized in that a parallel attention module is constructed, and because a large amount of background interference exists in a photovoltaic cell panel detection task, a parallel double-attention mechanism is adopted to optimize a U-Net network for feature extraction; the parallel double attention module comprises the following three parts: channel attention module, spatial attention module, feature fusion.
6. The photovoltaic cell panel inspection method based on deep learning of claim 1 is characterized in that a built model is used for training on the data set constructed in the step 1; for input image data X, obtaining a characteristic diagram after encoding processingThe calculation formula is as follows:
in the formula: conv stands for convolution operation; function C represents the feature aggregation mechanism: convolution layer + normalization; the Down function represents Down-sampling; l meterShowing the number of network layers; then decoding the obtained characteristic diagram to obtain the characteristic diagramThe calculation formula is as follows:
in the formula: [] Representing a concatenation of dimensions; up represents the Up-sampling of the function; bp represents a parallel attention mechanism operation; and (3) carrying out Sigmoid on the finally obtained feature graph to obtain a final prediction result:
the loss function is defined by a binary cross entropy loss function, defined as follows:
7. the photovoltaic cell panel inspection method based on deep learning of claim 5, wherein the channel attention module functions as follows:
firstly, defining the feature graph of an input module as F, and then obtaining a 1 × 1 × C global feature F after passing through a global maximum pooling layer1(ii) a Then using a one-dimensional convolution of size k on the global feature F1Performing convolution to obtain an attention weight map G; attention weight graph G weight mu of ith channeliCan be obtained by the following formula:
in the formula:j is more than or equal to 1 and less than or equal to k and represents the jth weight used for learning the ith channel weight;representing a set of global features of a k-th adjacent channel in the ith channel;
the convolution kernel size k of the one-dimensional convolution is calculated by:
in the formula: c is the number of channels; γ and b represent hyper-parameters, typically γ ═ 2, b ═ 1; | A |oddRepresents the nearest odd number adjacent to a; weighting the feature map F through the weight map G to obtain a feature map R; the weight of the channel related to the damage area to be detected is improved in the characteristic diagram R, so that the weight of other channels is reduced.
8. The photovoltaic cell panel inspection method based on deep learning of claim 5, wherein the spatial attention module functions as follows:
and respectively performing average pooling and maximum pooling on the feature map F in a channel dimension to generate two feature maps of a single channel: characteristic graph of mean valueAnd maximum value feature mapThen, combining the feature graphs of the two single channels to generate a weight graph H and weighting the feature graph F to generate a feature graph S;
the spatial attention module calculates as follows:
9. The photovoltaic cell panel inspection method based on deep learning of claim 5, wherein the feature fusion function is as follows:
adding the characteristic diagram R and the characteristic diagram S, and then obtaining a weight diagram G by using a ReLu activation function; the obtained weight graph is respectively fused with the weight distribution of the channel dimension and the weight distribution of the space dimension, so that more complementary features are obtained, and the feature regions such as damage, cracks and the like are concerned.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111113670.0A CN114021741A (en) | 2021-09-23 | 2021-09-23 | Photovoltaic cell panel inspection method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111113670.0A CN114021741A (en) | 2021-09-23 | 2021-09-23 | Photovoltaic cell panel inspection method based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114021741A true CN114021741A (en) | 2022-02-08 |
Family
ID=80054690
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111113670.0A Pending CN114021741A (en) | 2021-09-23 | 2021-09-23 | Photovoltaic cell panel inspection method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114021741A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115115610A (en) * | 2022-07-20 | 2022-09-27 | 南京航空航天大学 | Industrial CT (computed tomography) method for identifying internal defects of composite material based on improved convolutional neural network |
CN116596999A (en) * | 2023-04-19 | 2023-08-15 | 大连工业大学 | Automatic positioning method for pig carcass backfat thickness measurement position by combining deep learning and image processing |
CN117233615A (en) * | 2023-11-10 | 2023-12-15 | 中油绿电新能源有限公司 | Battery charging process abnormality detection method and device based on comparison learning network |
CN117556379A (en) * | 2024-01-12 | 2024-02-13 | 西南石油大学 | Photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint |
-
2021
- 2021-09-23 CN CN202111113670.0A patent/CN114021741A/en active Pending
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115115610A (en) * | 2022-07-20 | 2022-09-27 | 南京航空航天大学 | Industrial CT (computed tomography) method for identifying internal defects of composite material based on improved convolutional neural network |
CN115115610B (en) * | 2022-07-20 | 2023-08-22 | 南京航空航天大学 | Industrial CT composite material internal defect identification method based on improved convolutional neural network |
CN116596999A (en) * | 2023-04-19 | 2023-08-15 | 大连工业大学 | Automatic positioning method for pig carcass backfat thickness measurement position by combining deep learning and image processing |
CN116596999B (en) * | 2023-04-19 | 2024-04-05 | 大连工业大学 | Automatic positioning method for pig carcass backfat thickness measurement position by combining deep learning and image processing |
CN117233615A (en) * | 2023-11-10 | 2023-12-15 | 中油绿电新能源有限公司 | Battery charging process abnormality detection method and device based on comparison learning network |
CN117233615B (en) * | 2023-11-10 | 2024-02-06 | 中油绿电新能源有限公司 | Battery charging process abnormality detection method and device based on comparison learning network |
CN117556379A (en) * | 2024-01-12 | 2024-02-13 | 西南石油大学 | Photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint |
CN117556379B (en) * | 2024-01-12 | 2024-04-09 | 西南石油大学 | Photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114021741A (en) | Photovoltaic cell panel inspection method based on deep learning | |
CN113486865B (en) | Power transmission line suspended foreign object target detection method based on deep learning | |
CN112507793A (en) | Ultra-short-term photovoltaic power prediction method | |
CN113935977A (en) | Solar cell panel defect generation method based on generation countermeasure network | |
CN115409823A (en) | Solar screen defect detection method based on generative countermeasure network | |
CN115410078A (en) | Low-quality underwater image fish target detection method | |
CN108008633A (en) | Irradiation level comprising a variety of Changes in weather and photovoltaic module coordinate incidence relation method for building up | |
Hou et al. | Classification of defective photovoltaic modules in imagenet-trained networks using transfer learning | |
Cao et al. | Improved YOLOv8-GD deep learning model for defect detection in electroluminescence images of solar photovoltaic modules | |
CN115880691B (en) | Roof photovoltaic potential estimation method based on computer vision | |
CN116246060A (en) | Transmission line bolt defect detection method based on context reasoning | |
CN116612343A (en) | Power transmission line hardware detection method based on self-supervision learning | |
CN115049856A (en) | Fan blade fault detection method and system based on improved YOLOv5 | |
CN113496210B (en) | Photovoltaic string tracking and fault tracking method based on attention mechanism | |
CN115564100A (en) | Photovoltaic power prediction method, system and equipment | |
Liquan et al. | Fast detection of defective insulator based on improved YOLOv5s | |
CN112561181A (en) | Photovoltaic power generation prediction system based on Unet network and foundation cloud picture | |
Jiang et al. | An enhancement generative adversarial networks based on feature moving for solar panel defect identification | |
Cao et al. | IDS-Net: Integrated Network for Identifying Dust State of Photovoltaic Panels | |
Xi et al. | Defect Detection Algorithm of Photovoltaic Module EL Image Based on Two-stage YOLOts | |
CN113409237A (en) | Novel solar cell panel hot spot detection method based on YOLOv3 | |
Wang et al. | Photovoltaic Panel Intelligent Detection Method Based on Improved Faster-RCNN | |
CN117612029B (en) | Remote sensing image target detection method based on progressive feature smoothing and scale adaptive expansion convolution | |
CN116485802B (en) | Insulator flashover defect detection method, device, equipment and storage medium | |
Sun | Improved YOLOv7-based photovoltaic panel defect detection |
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 |