CN113496210B - Photovoltaic string tracking and fault tracking method based on attention mechanism - Google Patents
Photovoltaic string tracking and fault tracking method based on attention mechanism Download PDFInfo
- Publication number
- CN113496210B CN113496210B CN202110687502.6A CN202110687502A CN113496210B CN 113496210 B CN113496210 B CN 113496210B CN 202110687502 A CN202110687502 A CN 202110687502A CN 113496210 B CN113496210 B CN 113496210B
- Authority
- CN
- China
- Prior art keywords
- string
- fault
- tracking
- feature
- feature map
- 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.)
- Active
Links
- 230000007246 mechanism Effects 0.000 title claims abstract description 54
- 238000000034 method Methods 0.000 title claims abstract description 47
- 230000004927 fusion Effects 0.000 claims abstract description 42
- 230000008569 process Effects 0.000 claims abstract description 19
- 238000000605 extraction Methods 0.000 claims abstract description 11
- 238000010586 diagram Methods 0.000 claims description 33
- 230000002787 reinforcement Effects 0.000 claims description 12
- 230000009471 action Effects 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 6
- 238000005259 measurement Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000004891 communication Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000001514 detection method Methods 0.000 abstract description 8
- 238000012423 maintenance Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 208000033793 Neuroendocrine tumor of stomach Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
- 208000036506 well differentiated low or intermediate grade gastric neuroendocrine tumor Diseases 0.000 description 1
Classifications
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a photovoltaic string tracking and fault tracking method based on an attention mechanism, which comprises the steps of inputting a target frame into a pre-trained CNN network tracking model, outputting target tracking information of a photovoltaic string, numbering the target tracking information, outputting fault tracking information in the string, wherein the CNN network tracking model comprises three modules of feature extraction, feature fusion and prediction, and a feature extraction part utilizes CNN high-dimensional and low-dimensional information to assist feature extraction; the feature fusion component increases the attention channel and the generalized generic GNet channel. Because the feature extraction channel is added with high-dimensional and low-dimensional features, the robustness in the motion process is improved; meanwhile, as the GNet channel, the context enhancement module and the cross feature enhancement module are introduced into the feature fusion part, semantic features are fully considered in the tracking process, the robustness of the attributes such as motion change and the like in the tracking process is improved, and accurate results are obtained when photovoltaic string tracking and fault detection tracking are carried out.
Description
Technical Field
The invention belongs to the technical field of computer digital image processing, and particularly relates to a photovoltaic group string tracking and fault tracking method based on an attention mechanism.
Background
At present, photovoltaic power generation is a representative technology in the field of new energy, and is widely applied at home and abroad. Considering the environmental difference of the distribution of the photovoltaic installation in China, the photovoltaic power generation panel often causes the problem of hot spots due to external factors such as weather and environment, and the like, so that local power transmission faults can be caused, the power transmission efficiency is affected, and even the solar module is burnt.
With the development of image processing technology, tracking and detection technologies based on deep learning are also becoming better and better represented. Compared with the traditional method, the deep learning-based photovoltaic string tracking and fault tracking method has the advantages that the target tracking is applied to the photovoltaic field, the photovoltaic management and fault detection efficiency is improved, and personnel maintenance is facilitated.
For existing fault detection and photovoltaic management work, an infrared image fault detection method or a traditional manual inspection method is adopted in most cases, and for large-scale panel management and fault inspection work, high efficiency and high accuracy cannot be met at the same time, so that the photovoltaic equipment cannot be managed in time, and serious energy waste and potential safety hazards are caused.
Disclosure of Invention
The invention aims to provide a photovoltaic string tracking and fault tracking method based on an attention mechanism, which can accurately describe string and fault information in the string, and improve detection accuracy.
The technical scheme adopted by the invention is that the photovoltaic string tracking and fault tracking method based on the attention mechanism is implemented according to the following steps:
step 1, taking a plurality of groups of photovoltaic string pictures as samples, and training a CNN network tracking model based on an attention mechanism;
step 2, obtaining a target video, inputting the target video into a trained CNN network tracking model based on an attention mechanism to obtain a video frame sequence, extracting features of the video frame sequence, and outputting a string feature map and a fault feature map of the video frame sequence;
step 3, combining an attention mechanism, carrying out feature reinforcement on the string feature map, removing noise interference, carrying out reinforcement fusion on the fault feature map, and inputting the denoised string feature map into a GNet network to obtain a thermodynamic diagram;
step 4, inputting the denoised string feature images and the fault feature images subjected to enhancement fusion into a cross feature enhancement module CFA, performing attention-based cross fusion, and outputting the feature images subjected to fusion;
step 5, inputting the fused feature images into a pre-measurement head, and carrying out classification and regression through the pre-measurement head to obtain a fault foreground/background classification result;
and 6, outputting the fault type and the fault position of the photovoltaic group string according to the thermodynamic diagram and the classification result.
The invention is also characterized in that:
the specific process of the step 2 is as follows:
the method comprises the steps of collecting a video of a photovoltaic string through an unmanned aerial vehicle as a target video, inputting the target video into a trained CNN network tracking model based on an attention mechanism, extracting string characteristics through a high-dimensional characteristic extraction channel con-5 in the CNN based on the attention mechanism to obtain a string characteristic diagram, and selecting and extracting internal fault detail information of the string through a low-dimensional channel con-4 to obtain the fault characteristic diagram.
And 3, carrying out characteristic reinforcement on the string characteristic diagram and carrying out reinforcement fusion on the fault characteristic diagram, wherein the specific process comprises the following steps of: and combining an attention mechanism, performing feature enhancement on the string feature map or the fault feature map by using two context enhancement modules ECA, wherein ECA is multi-head self-attention in a residual form, adaptively integrating information of different positions in the string feature map or the fault feature map, taking the string feature map or the fault feature map as input of the ECA mechanism, and obtaining a denoised string feature map or an enhanced fused fault feature map, wherein the action mechanism of ECA expresses:
X EC ＝X+MultiHead(X+P x ,X+P x ,X) (1)
wherein X represents the input of ECA, p x For spatial position coding, X EC Is the output of ECA.
In step 4, spatial position coding is adopted in the cross feature enhancement module CFA, and an FFN module is cited in the cross feature enhancement module CFA, wherein the FFN module is a full-communication feed-forward network consisting of two linear transforms, and one ReLU is arranged between the two linear transforms, namely
FFN(x)＝max(0,xW 1 +b 1 )W 2 +b 2 (2)
Where the symbol w represents a weight matrix, b represents a basis vector, and the subscript represents the different layers of the fusion.
The specific process of the step 4 is as follows: inputting the denoised string feature map and the enhanced and fused fault feature map into a cross feature enhancement module CFA, repeating the fusion part for N times under the action mechanism of the CFA, and outputting the fused feature map;
among them, the mechanism of CFA action is summarized as:
wherein X is q Is the vector representation form of the fault characteristic diagram after enhanced fusion, P q Is corresponding to X q Spatial position coding of X kv Is the vector representation form of the denoised string characteristic diagram, p kv Is X kv The spatial encoding of the coordinates is performed,is output through fusion of multi-head cross features, X CF Is the output of the CFA.
In step 5, the prediction head includes a classification branch and a regression branch, and the classification branch and the regression branch form three-layer perceptrons with hidden dimension d and ReLU activation function.
The specific process of the step 5 is as follows: inputting the fused feature images into a pre-measuring head, classifying the feature images through binary cross entropy loss in a classification branch, and performing regression training on a regression branch to obtain a fault prospect/background classification result;
wherein, the classification branch is defined as:
L cls ＝-∑[y i log(p j )+(1-y i )log(1-p j )] (4)
y j group-trunk tag representing jth sample, y j =1 denotes foreground, p j Representing the probability of belonging to the foreground predicted by the learning model;
the definition of regression loss in the regression branch is:
y i =1 meansPositive sample, b j Representing the jth prediction bounding box,representing normalized real bounding boxes, lambda G And lambda (lambda) 1 Is a regularization parameter.
The specific process of the step 6 is as follows: obtaining a predicted position of the photovoltaic string according to the thermodynamic diagram, and tracking the serial number of the string according to the predicted position of the photovoltaic string; and 5, obtaining a fault foreground/background classification result according to the step, and outputting a classification and tracking result of faults in the group string.
The invention has the beneficial effects that:
according to the photovoltaic string tracking and fault tracking method based on the attention mechanism, the number tracking of the photovoltaic string is completed by combining the feature fusion of the attention mechanism through the features of different layers of the pre-trained CNN tracking network based on the attention mechanism, meanwhile, the fault tracking output of the string inside is completed, the string and the fault information inside the string can be accurately described, the detection accuracy is improved, the operation and maintenance cost is reduced, and the maintenance of workers is facilitated.
Drawings
FIG. 1 is a schematic diagram of the overall network structure of the photovoltaic string tracking and fault tracking method based on the attention mechanism of the present invention;
FIG. 2 is a schematic flow chart of a photovoltaic string tracking and fault tracking method based on an attention mechanism;
FIG. 3 is a schematic diagram of a feature fusion portion context enhancement module (ECA) provided by the present invention;
FIG. 4 is a schematic diagram of a feature fusion portion cross feature enhancement module (CFA) provided by the present invention;
FIG. 5 is an original image of a photovoltaic string acquired in an embodiment of the present invention;
FIG. 6 is an effect diagram of group string number tracking provided by an embodiment of the present invention;
fig. 7 is a string number and fault location diagram of a photovoltaic string tracking and fault tracking method employing an attention-based mechanism in an embodiment of the present invention.
Detailed Description
The invention will be described in detail below with reference to the drawings and the detailed description.
The invention provides a photovoltaic string tracking and fault tracking method based on an attention mechanism, wherein a network of the method is shown in figure 1, and mainly comprises three parts of feature extraction, feature fusion and target prediction, so that the string and fault information inside the string can be accurately described, and the serial number tracking of the photovoltaic string and the accurate output of the fault information are realized.
The feature extraction part combines the pre-trained CNN features of different levels based on the attention mechanism, wherein the high dimension contains more semantic information and can be used for extracting a string feature map; the low dimension contains more local features that will help extract faults from the string.
And the feature fusion part is combined with the attention mechanism, and firstly utilizes two context enhancement modules (ECA) to respectively conduct feature enhancement on the group string and the fault feature graph so as to respectively strengthen the group string and the fault feature. And then, the cross feature enhancement (CFA) module is utilized to carry out weighted fusion on the group string feature map and the fault feature map in a plurality of scales after ECA enhancement so as to enhance the representation capability of the feature map. For the string feature, sending the string feature into the GNet network for generating a string thermodynamic diagram; for fault features, the group string feature map and the fault feature map are weighted and fused in multiple scales by utilizing a cross feature enhancement (CFA) module to enhance the representation capability of the feature map.
And a prediction part, which obtains the predicted string position according to the string thermodynamic diagram, so as to carry out tracking numbering of the string. And for the fault fusion feature map, a prediction head comprising a classification branch and a prediction branch is adopted, and the feature map is classified and regressed, so that a foreground/background classification result and the normalized coordinates of a fault prediction area are obtained, and the fault positioning output work is completed.
The invention discloses a photovoltaic group string tracking and fault tracking method based on an attention mechanism, which is implemented according to the following steps as shown in fig. 2:
step 1, taking a plurality of groups of photovoltaic string pictures as samples, and training a CNN network tracking model based on an attention mechanism;
step 2, obtaining a target video, inputting the target video into a trained CNN network tracking model based on an attention mechanism, reading and obtaining a video frame sequence according to frames, extracting features of the video frame sequence, and outputting a string feature map and a fault feature map of the video frame sequence; the specific process of the step 2 is as follows:
collecting video of a photovoltaic string as a target video through an unmanned aerial vehicle, inputting the target video into a trained CNN network tracking model based on an attention mechanism, and extracting string characteristics through a high-dimensional characteristic extraction channel con-5 (CNN layer 15) in a CNN network based on the attention mechanism to obtain a string characteristic diagram f z The internal fault detail information of the group string is extracted through the target image feature selection by the low-dimensional channel con-4 (CNN layer 10), so that faults with similar appearance can be better distinguished, and a fault feature map f is obtained x 。
Step 3, combining an attention mechanism, carrying out feature reinforcement on the string feature map, removing noise interference, carrying out reinforcement fusion on the fault feature map, inputting the denoised string feature map into a GNet network, wherein the GNet network is a CNN design about target positioning, constructing by utilizing string features of a first frame, and obtaining a thermodynamic diagram for capturing position, category and contour information of the string by a network formed by convolution kernel size of a convolution layer of 5 multiplied by 5;
step 4, inputting the denoised string feature images and the fault feature images subjected to enhancement fusion into a cross feature enhancement module CFA, performing attention-based cross fusion, and outputting the feature images subjected to fusion;
and 3, carrying out characteristic reinforcement on the string characteristic diagram, and carrying out reinforcement fusion on the fault characteristic diagram in the step 4, wherein the specific process comprises the following steps: the method comprises the steps of combining an attention mechanism, utilizing two context enhancement modules ECA to perform feature enhancement on a string feature map or a fault feature map, wherein ECA is multi-head self-attention in a residual form, and adaptively integrating information of different positions in the string feature map or the fault feature map, and considering that the attention mechanism does not have distinguishing capability on input feature positions, so that a spatial position coding process is combined. Taking the string feature map or the fault feature map as input of an ECA mechanism to obtain a denoised string feature map or an enhanced fused fault feature map, wherein the ECA mechanism expresses:
X EC ＝X+MultiHead(X+P x ,X+P x ,X) (1)
wherein X represents the input of ECA, p x For spatial position coding, X EC Is the output of ECA.
In step 4, spatial position coding is adopted in the cross feature enhancement module CFA, and an FFN module is cited in the cross feature enhancement module CFA, wherein the FFN module is a full-communication feed-forward network consisting of two linear transforms, and one ReLU is arranged between the two linear transforms, namely
FFN(x)＝max(0,xW 1 +b 1 )W 2 +b 2 (2)
Where the symbol w represents a weight matrix, b represents a basis vector, and the subscript represents the different layers of the fusion.
The specific process of the step 4 is as follows: inputting the denoised string feature map and the enhanced and fused fault feature map into a cross feature enhancement module CFA, repeating the fusion part for 4 times under the action mechanism of the CFA, and outputting the fused feature map;
among them, the mechanism of CFA action is summarized as:
wherein X is q Is the vector representation form of the fault characteristic diagram after enhanced fusion, P q Is corresponding to X q Spatial position coding of X kv Is the vector representation form of the denoised string characteristic diagram, p kv Is X kv The spatial encoding of the coordinates is performed,is output through fusion of multi-head cross features, X CF Is the output of the CFA. According to the above formula, CFA is according to X kv And X q Multiple scales between them calculate an attention map and then calculate an attention map for X based on the attention map kv Re-addingWeight and add it to X q To enhance the representational capacity of the feature map.
Step 5, inputting the fused feature images into a pre-measuring head, and predicting each vector through the pre-measuring head so as to obtain a foreground/background classification result and a normalized coordinate representing the size of a search area; the prediction head comprises a classification branch and a regression branch, and the classification branch and the regression branch form three-layer perceptrons with hidden dimension d and ReLU activation functions.
The specific process of the step 5 is as follows: inputting the fused feature images into a pre-measuring head, classifying the feature images through binary cross entropy loss in a classification branch, and performing regression training on a regression branch to obtain a fault prospect/background classification result;
wherein, the classification branch is defined as:
L cls ＝-∑[y i log(p j )+(1-y i )log(1-p j )] (4)
y j group-trunk tag representing jth sample, y j =1 denotes foreground, p j Representing the probability of belonging to the foreground predicted by the learning model;
the definition of regression loss in the regression branch is:
y i =1 denotes positive sample, b j Representing the jth prediction bounding box,representing normalized real bounding boxes, lambda G And lambda (lambda) 1 Is a regularization parameter.
And 6, outputting the fault type and the fault position of the photovoltaic group string according to the thermodynamic diagram and the classification result. The specific process is as follows: obtaining a predicted position of the photovoltaic string according to the thermodynamic diagram, and tracking the serial number of the string according to the predicted position of the photovoltaic string; and 5, obtaining a fault foreground/background classification result according to the step, and outputting a classification and tracking result of faults in the group string.
Examples
For the photovoltaic power generation industry in the new energy field, the unmanned aerial vehicle is combined for shooting, the photovoltaic system of a distributed type (factory roof or resident roof) is verified, the method is utilized for tracking 10 groups of 33 m-height photovoltaic group series videos, each video is 255 frames, the original image of the collected photovoltaic group series is shown in fig. 5, a white area with a serial number being selected as a group series, a small square in the group series is called a panel, faults are generally divided into group series faults (the whole group series is faults), panel faults, hot plate faults (bright spots in the panel) and strip faults (bright strips in the panel), the characteristic extraction and characteristic fusion of the group series are carried out, the thermodynamic diagram of the group series is obtained by combining a generalized GNET network, the result prediction of the group series is completed according to the thermodynamic diagram, the effect is shown in fig. 6, and the tracking number information of the group series can be obtained according to fig. 6. Obtaining a fusion vector of fault information in the group string through feature extraction, context feature fusion and background information crossing group string and fault feature fusion of fault faults, completing identification and positioning of fault information in the group string by combining classification and regression in the step 5, outputting fault position information (comprising corresponding group string numbers, fault positions (x_starting point horizontal coordinates, y_vertical coordinates, w_fault width and h_fault height)) in the group string, completing identification of faults, marking on frames, and achieving the effect of figure 7 in the attached part, thereby completing tracking of the group string numbers and the fault tracking in the group string, wherein the accuracy is not lower than 95%; through the data, the method provided by the invention is verified, so that the management of the group string can be facilitated, and on the other hand, more accurate information about faults can be obtained, and the maintenance of the photovoltaic is facilitated.
Through the method, the photovoltaic string tracking and fault tracking method based on the attention mechanism, the number tracking of the photovoltaic string is completed by combining the feature fusion of the attention mechanism through the features of different layers of the pre-trained CNN tracking network based on the attention mechanism, meanwhile, the fault tracking output of the string inside is completed, the string and the fault information inside the string can be accurately described, the detection accuracy is improved, the operation and maintenance cost is reduced, and the maintenance of workers is facilitated.
Claims (5)
1. The photovoltaic string tracking and fault tracking method based on the attention mechanism is characterized by comprising the following steps of:
step 1, taking a plurality of groups of photovoltaic string pictures as samples, and training a CNN network tracking model based on an attention mechanism;
step 2, obtaining a target video, inputting the target video into a trained CNN network tracking model based on an attention mechanism to obtain a video frame sequence, extracting features of the video frame sequence, and outputting a string feature map and a fault feature map of the video frame sequence;
step 3, combining an attention mechanism, carrying out feature reinforcement on the string feature map, removing noise interference, carrying out reinforcement fusion on the fault feature map, and inputting the denoised string feature map into a GNet network to obtain a thermodynamic diagram;
the specific process of carrying out feature reinforcement on the string feature map and carrying out reinforcement fusion on the fault feature map is as follows: and combining an attention mechanism, performing feature enhancement on the string feature map or the fault feature map by using two context enhancement modules ECA, wherein ECA is multi-head self-attention in a residual form, adaptively integrating information of different positions in the string feature map or the fault feature map, taking the string feature map or the fault feature map as input of the ECA mechanism, and obtaining a denoised string feature map or an enhanced fused fault feature map, wherein the action mechanism of ECA expresses:
X EC ＝X+MultiHead(X+P x ,X+P x ,X) (1)
wherein X represents the input of ECA, p x For spatial position coding, X EC Is the output of ECA;
step 4, inputting the denoised string feature images and the fault feature images subjected to enhancement fusion into a cross feature enhancement module CFA, performing attention-based cross fusion, and outputting the feature images subjected to fusion; the specific process is as follows: inputting the denoised string feature map and the enhanced and fused fault feature map into a cross feature enhancement module CFA, repeating the fusion part for N times under the action mechanism of the CFA, and outputting the fused feature map;
among them, the mechanism of CFA action is summarized as:
wherein X is q Is the vector representation form of the fault characteristic diagram after enhanced fusion, P q Is corresponding to X q Spatial position coding of X kv Is the vector representation form of the denoised string characteristic diagram, p kv Is X kv The spatial encoding of the coordinates is performed,is output through fusion of multi-head cross features, X CF Is the output result of the CFA;
spatial position coding is adopted in the cross feature enhancement module CFA, and an FFN module is cited in the cross feature enhancement module CFA, wherein the FFN module is a full-communication feed-forward network consisting of two linear transforms, and one ReLU is arranged between the two linear transforms, namely
FFN(x)＝max(0,xW 1 +b 1 )W 2 +b 2 (2)
Wherein the symbol w represents a weight matrix, b represents a base vector, and the subscript represents different layers of fusion;
step 5, inputting the fused feature images into a pre-measurement head, and carrying out classification and regression through the pre-measurement head to obtain a fault foreground/background classification result;
and 6, outputting the fault type and the fault position of the photovoltaic group string according to the thermodynamic diagram and the classification result.
2. The method for tracking photovoltaic strings and faults based on an attention mechanism according to claim 1, wherein the specific process of the step 2 is as follows:
the method comprises the steps of collecting a video of a photovoltaic string through an unmanned aerial vehicle as a target video, inputting the target video into a trained CNN network tracking model based on an attention mechanism, extracting string characteristics through a high-dimensional characteristic extraction channel con-5 in the CNN based on the attention mechanism to obtain a string characteristic diagram, and selecting and extracting internal fault detail information of the string through a low-dimensional channel con-4 to obtain the fault characteristic diagram.
3. The method of claim 1, wherein the prediction head in step 5 includes a classification branch and a regression branch, and the classification branch and the regression branch each include a three-layer perceptron having hidden dimensions d and a ReLU activation function.
4. The method for tracking photovoltaic strings and faults based on an attention mechanism according to claim 3, wherein the specific process of the step 5 is as follows: inputting the fused feature images into a pre-measuring head, classifying the feature images through binary cross entropy loss in a classification branch, and performing regression training on a regression branch to obtain a fault prospect/background classification result;
wherein, the classification branch is defined as:
L cls ＝-∑[y i log(p j )+(1-y i )log(1-p j )] (4)
y j group-trunk tag representing jth sample, y j =1 denotes foreground, p j Representing the probability of belonging to the foreground predicted by the learning model;
the definition of regression loss in the regression branch is:
y i =1 denotes positive sample, b j Representing the jth prediction bounding box,representing normalized real bounding boxes, lambda G And lambda (lambda) 1 Is a regularization parameter.
5. The method for tracking photovoltaic strings and faults based on an attention mechanism according to claim 1, wherein the specific process of the step 6 is as follows: obtaining a predicted position of the photovoltaic string according to the thermodynamic diagram, and tracking the serial number of the string according to the predicted position of the photovoltaic string; and 5, obtaining a fault foreground/background classification result according to the step, and outputting a classification and tracking result of faults in the group string.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110687502.6A CN113496210B (en) | 2021-06-21 | 2021-06-21 | Photovoltaic string tracking and fault tracking method based on attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110687502.6A CN113496210B (en) | 2021-06-21 | 2021-06-21 | Photovoltaic string tracking and fault tracking method based on attention mechanism |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113496210A CN113496210A (en) | 2021-10-12 |
| CN113496210B true CN113496210B (en) | 2024-02-02 |
Family
ID=77997343
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110687502.6A Active CN113496210B (en) | 2021-06-21 | 2021-06-21 | Photovoltaic string tracking and fault tracking method based on attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113496210B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117520924B (en) * | 2023-12-29 | 2024-04-12 | 国网浙江省电力有限公司舟山供电公司 | Island photovoltaic operation and maintenance fault cause analysis method and system based on multi-mode data |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111192292A (en) * | 2019-12-27 | 2020-05-22 | 深圳大学 | Target tracking method based on attention mechanism and twin network and related equipment |
| CN112560656A (en) * | 2020-12-11 | 2021-03-26 | 成都东方天呈智能科技有限公司 | Pedestrian multi-target tracking method combining attention machine system and end-to-end training |
| CN112560695A (en) * | 2020-12-17 | 2021-03-26 | 中国海洋大学 | Underwater target tracking method, system, storage medium, equipment, terminal and application |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10740654B2 (en) * | 2018-01-22 | 2020-08-11 | Qualcomm Incorporated | Failure detection for a neural network object tracker |
-
2021
- 2021-06-21 CN CN202110687502.6A patent/CN113496210B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111192292A (en) * | 2019-12-27 | 2020-05-22 | 深圳大学 | Target tracking method based on attention mechanism and twin network and related equipment |
| CN112560656A (en) * | 2020-12-11 | 2021-03-26 | 成都东方天呈智能科技有限公司 | Pedestrian multi-target tracking method combining attention machine system and end-to-end training |
| CN112560695A (en) * | 2020-12-17 | 2021-03-26 | 中国海洋大学 | Underwater target tracking method, system, storage medium, equipment, terminal and application |
Non-Patent Citations (2)
| Title |
|---|
| 周双双 ; 宋慧慧 ; 张开华 ; 樊佳庆 ; .基于增强语义与多注意力机制学习的深度相关跟踪.计算机工程.2020,(第02期),全文. * |
| 齐天卉 ; 张辉 ; 李嘉锋 ; 卓力 ; .基于多注意力图的孪生网络视觉目标跟踪.信号处理.2020,(第09期),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113496210A (en) | 2021-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2019214268A1 (en) | Photovoltaic array fault diagnosis method based on composite information | |
| CN111368690B (en) | Deep learning-based video image ship detection method and system under influence of sea waves | |
| CN111696075A (en) | Intelligent fan blade defect detection method based on double-spectrum image | |
| CN112164038A (en) | Photovoltaic hot spot detection method based on deep convolutional neural network | |
| CN105825511A (en) | Image background definition detection method based on deep learning | |
| CN111444939A (en) | Small-scale equipment component detection method based on weak supervision cooperative learning in open scene of power field | |
| CN115205264A (en) | High-resolution remote sensing ship detection method based on improved YOLOv4 | |
| CN112700476A (en) | Infrared ship video tracking method based on convolutional neural network | |
| CN113496210B (en) | Photovoltaic string tracking and fault tracking method based on attention mechanism | |
| CN114529821A (en) | Offshore wind power safety monitoring and early warning method based on machine vision | |
| CN112668662B (en) | Outdoor mountain forest environment target detection method based on improved YOLOv3 network | |
| Cao et al. | Improved YOLOv8-GD deep learning model for defect detection in electroluminescence images of solar photovoltaic modules | |
| CN111091533B (en) | Battery piece EL defect detection method based on improved SSD algorithm | |
| CN111539950A (en) | Solar panel defect detection system based on machine vision | |
| CN114596273B (en) | Intelligent detection method for multiple defects of ceramic substrate by using YOLOV4 network | |
| CN110826478A (en) | Aerial photography illegal building identification method based on countermeasure network | |
| CN116309498A (en) | ResNeSt algorithm-based photovoltaic panel integrity detection method, system and storage medium | |
| CN113284103B (en) | Substation equipment defect online detection method based on space transformation fast R-CNN model | |
| CN113887443A (en) | Industrial smoke emission identification method based on attribute perception attention convergence | |
| Zheng et al. | Multiscale Fusion Network for Rural Newly Constructed Building Detection in Unmanned Aerial Vehicle Imagery | |
| Song et al. | PDD: Post-disaster Dataset for Human Detection and Performance Evaluation | |
| Kan et al. | Real-Time domestic garbage detection method based on improved YOLOv5 | |
| Yuan et al. | Smoky vehicle detection based on improved vision transformer | |
| Sheng et al. | A YOLOX-Based Detection Method of Triple-Cascade Feature Level Fusion for Power System External Defects | |
| Kang et al. | Efficient Object Detection with Deformable Convolution for Optical Remote Sensing Imagery |
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 |