Intelligent detection method and system for surface damage of photovoltaic panel
A neural network-based system for solar panel inspection accurately identifies and localizes damage, improving efficiency and reducing human error in solar panel maintenance.
Patent Information
- Application Number
- CN202510496647.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-21
- Publication Date
- 2025-07-15
AI Technical Summary
The prior art is difficult to efficiently identify and locate damage to the surface of photovoltaic panels, especially failures such as heat spots, hidden cracks and snail patterns, resulting in reduced power generation efficiency and shortened service life of photovoltaic modules, and low maintenance efficiency.
A damage detection model based on graph convolutional neural network is constructed. By collecting photovoltaic panel surface images and pre-processing, the damage detection model can be trained, so as to quickly identify and accurately locate the damage areas and categories of photovoltaic panels to reduce background noise interference.
It realizes rapid identification and precise positioning of surface damage of photovoltaic panels, improves photovoltaic power generation efficiency, and reduces losses caused by failures.
Smart Images
Figure CN120318208A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent detection of surface damage of photovoltaic panels, and particularly to an intelligent detection method and system for surface damage of photovoltaic panels. Background Art
[0002] In recent years, China's photovoltaic industry has witnessed great development, achieving high economic benefits, and the deployment of distributed photovoltaic equipment has been increasing. The most core component in the photovoltaic industry is the photovoltaic panel. However, the following faults are likely to occur in photovoltaic panels: (1) Hot spot problem, where burned hot spots appear on the panel surface, which can cause the entire photovoltaic module to be damaged, significantly reduce the power generation efficiency of the photovoltaic module, and shorten the service life of the photovoltaic module; (2) Cracking problem, where fine cracks appear in the photovoltaic module, generally caused by external forces, and are visible to the naked eye when reaching fragmentation, which has a greater impact on the output power of the photovoltaic module; (3) Snail trail problem, where black or white linear patterns appear on the photovoltaic panel surface, looking like the trails left by snails. Due to the increasingly widespread deployment of current photovoltaic power stations and various distributed photovoltaic equipment, the maintenance of photovoltaic panels by manual labor is inefficient, some minor surface damages are easily overlooked, and it is easily affected by environmental, weather, and human subjective factors.
[0003] Therefore, there is an urgent need for an intelligent detection method and system for surface damage of photovoltaic panels. Summary of the Invention
[0004] The purpose of the present invention is to provide an intelligent detection method and system for surface damage of photovoltaic panels. By collecting normal and damaged images of the photovoltaic panel surface and preprocessing them, the graph convolutional neural network is trained to construct a damage detection model, which can quickly identify the surface damage area and damage category of the photovoltaic panel, is applicable to the situation of multiple faults on the photovoltaic panel surface, can accurately locate the damage area, effectively avoid various background noise interferences caused by coupling, can improve the photovoltaic power generation efficiency, and reduce the losses caused by faults.
[0005] To achieve the above purpose, the present invention provides the following solutions:
[0006] An intelligent detection method for surface damage of photovoltaic panels includes:
[0007] Obtain the surface image of the photovoltaic panel to be detected;
[0008] Input the surface image into a preset damage detection model, and output the surface damage area and damage category of the photovoltaic panel to be detected; wherein, the damage detection model is obtained by training based on a training set, the training set includes a number of normal pictures of the photovoltaic panel surface and damaged pictures of the photovoltaic panel surface, and the damage detection model is constructed by a graph neural network.
[0009] Optionally, before training the damage detection model based on the training set, preprocessing the training set is also included. The preprocessing includes:
[0010] Cleaning, screening, labeling the damage areas and damage categories of the collected normal pictures of the photovoltaic panel surface and damaged pictures of the photovoltaic panel surface, and then performing color correction, random rotation, and random noise addition to complete the preprocessing.
[0011] Optionally, the damage detection model includes: a layer-by-layer feature extraction module, a global feature extraction module, and a damage discrimination module. Among them, the layer-by-layer feature extraction module is used to extract damage features at different levels from the surface image; the global feature extraction module is used to obtain global damage features based on damage features at different levels; the damage discrimination module is used to judge the damage category and damage area based on the global damage features.
[0012] Optionally, the layer-by-layer feature extraction module includes: a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, and a fourth feature extraction layer. Among them, the first feature extraction layer preprocesses the surface image through 7×7 convolution, batch standard normalization, and max pooling operations; the second feature extraction layer, the third feature extraction layer, and the fourth feature extraction layer respectively perform feature extraction on the surface image processed by the first feature extraction layer through different numbers of 3×3 convolutions and 1×1 convolutions, obtain damage feature vectors and damage category vectors of different dimensions and splice them, and output a damage area feature space vector.
[0013] Optionally, the global feature extraction module's extraction of damage features at different levels from the surface image includes:
[0014] After normalizing the damage area feature space vector through layer standard normalization, then using an attention mechanism to learn the damage attention weights of different damage areas;
[0015] Combining the damage area feature space vector and the damage attention weights of different damage areas to obtain the global damage feature.
[0016] Optionally, the damage discrimination module's judgment of the damage category and damage area based on the global damage features includes:
[0017] Inputting the global damage feature into two fully connected layers and a Softmax classification layer in sequence, and outputting the damage category and damage area of the photovoltaic panel to be detected.
[0018] Optionally, during the process of training the damage detection model, it is judged that the training of the damage detection model ends when the loss function converges or the iteration reaches a preset number of times. Among them, the loss function is:
[0019]
[0020] In the formula, L is the cross-entropy loss, t is the number of samples, and y (i) is the actual sample label, is the label discriminated by the Softmax layer.
[0021] To further achieve the above object, the present invention also provides an intelligent detection system for surface damage of a photovoltaic panel, including: an image acquisition module and a damage detection module. Among them, the image acquisition module is used to collect the surface image of the photovoltaic panel to be detected based on an image acquisition device carried by a drone; the damage detection module is used to input the surface image into a damage detection model and output the surface damage area and damage category of the photovoltaic panel to be detected. Among them, the damage detection model includes a layer-by-layer feature extraction module, a global feature extraction module, and a damage discrimination module. The layer-by-layer feature extraction module is used to extract damage features at different levels from the surface image; the global feature extraction module is used to obtain global damage features based on damage features at different levels; the damage discrimination module is used to judge the damage category and damage area based on the global damage features.
[0022] The beneficial effects of the present invention are as follows:
[0023] By collecting the normal and damaged images of the surface of the photovoltaic panel, preprocessing them, and then training the graph convolutional neural network, the present invention constructs a damage detection model, which can quickly identify the surface damage area and damage category of the photovoltaic panel, is applicable to the situation of multiple faults on the surface of the photovoltaic panel, can accurately locate the damage area, effectively avoid various background noise interferences of coupling, can improve the power generation efficiency of photovoltaic power generation, and reduce the losses caused by faults. Description of the Drawings
[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.
[0025] Figure 1 It is a flowchart of an intelligent detection method for surface damage of a photovoltaic panel according to an embodiment of the present invention. Detailed Embodiments
[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0027] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.
[0028] This embodiment provides an intelligent detection method for surface damage of a photovoltaic panel, as Figure 1 shown, including:
[0029] Obtain the surface image of the photovoltaic panel to be detected;
[0030] Input the surface image into a preset damage detection model, and output the surface damage area and damage category of the photovoltaic panel to be detected; wherein, the damage detection model is obtained by training based on a training set, the training set includes a number of normal pictures of the photovoltaic panel surface and damaged pictures of the photovoltaic panel surface, and the damage detection model is constructed by a graph neural network.
[0031] Further, the surface image of the photovoltaic panel to be detected is obtained by an image acquisition device and a ranging sensor carried by a drone when the drone is relatively stationary with respect to the photovoltaic panel to be detected and is located on the central line of the photovoltaic panel to be detected.
[0032] Further, before training the damage detection model based on the training set, preprocessing of the training set is also included. The preprocessing includes:
[0033] Clean, screen, label the damage area and damage category of the collected normal pictures of the photovoltaic panel surface and damaged pictures of the photovoltaic panel surface, and then perform color correction, random rotation, and random noise addition to complete the preprocessing.
[0034] Specifically, first screen and clean the collected picture samples to obtain clear sample pictures, and then label the damage area and damage category of the sample pictures according to whether there is damage to the photovoltaic panel in the sample pictures. The damage area is presented in the form of a square frame, and the damage categories include no damage, hot spot, hidden crack, snail pattern, etc.; then expand the collected picture samples, and construct the final training set through methods such as color correction, random rotation, and random noise addition.
[0035] In this embodiment, the random rotation angles are set to 60°, 120°, and 240°; the random noise addition is to add noise b to the corresponding pixel point a in the pixel matrix of the sample picture:
[0036] c(x,y) = a(x,y) + b(x,y)
[0037] Further, the damage detection model includes: a layer-by-layer feature extraction module, a global feature extraction module, and a damage discrimination module. Among them, the layer-by-layer feature extraction module is used to extract damage features at different levels from the surface image; the global feature extraction module is used to obtain global damage features based on the damage features at different levels; the damage discrimination module is used to judge the damage category and damage area based on the global damage features.
[0038] Further, the layer-by-layer feature extraction module includes: a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, and a fourth feature extraction layer. Among them, the first feature extraction layer preprocesses the surface image through 7×7 convolution, batch standard normalization, and max pooling operations; the second feature extraction layer, the third feature extraction layer, and the fourth feature extraction layer respectively perform feature extraction on the surface image processed by the first feature extraction layer through different numbers of 3×3 convolutions and 1×1 convolutions, obtain damage feature vectors and damage category vectors of different dimensions and splice them, and output a damage area feature space vector.
[0039] Specifically, in this embodiment, the damage detection model is constructed based on the ResNet network, extracts local convolution features of the surface image in stages, quantifies the number of channels of the feature vector through the convolution layer, then performs linear flattening and splices it with the damage category vector to construct a feature space vector of the surface damage area of the photovoltaic panel to be detected.
[0040] The first feature extraction layer undergoes 64 convolution operations of size 7×7 and BatchNorm normalization, and then passes through a 2×2 MaxPool layer to preprocess the image; the second feature extraction layer is mainly composed of three residual modules, which initially extract the image damage features. The input vector is mainly subjected to feature extraction through 64 convolution layers of size 3×3, and then the number of channels is quantified through 256 convolution layers of size 1*1; the third feature extraction layer and the fourth feature extraction layer have a similar structure to the second feature extraction layer, mainly used for deeper feature extraction. Assuming the original image size is H×W×C, where H and W represent the resolution of the image, and C represents the number of channels, after three stages of convolution feature extraction, the output dimension is (H / 16)×(W / 16)×1024; the number of channels of the feature vector is adjusted through 768 convolution layers of size 1*1, the vector is linearly flattened to (HW / 256)×1024, and the feature vector is spliced with the damage category vector to construct a feature space vector of the surface damage area of the photovoltaic panel.
[0041] Further, the global feature extraction module's extraction of damage features at different levels from the surface image includes:
[0042] After normalizing the feature space vector of the damaged area through layer standard normalization, the damage attention weights of different damaged areas are learned using the attention mechanism;
[0043] Combining the feature space vector of the damaged area and the damage attention weights of different damaged areas to obtain the global damage feature.
[0044] Furthermore, the damage discrimination module determines the damage category based on the global damage feature. The damaged areas include:
[0045] The global damage feature is sequentially input into two fully connected layers and a Softmax classification layer to output the damage category and damaged area of the photovoltaic panel to be detected.
[0046] Furthermore, during the process of training the damage detection model, it is judged that the training of the damage detection model is completed when the loss function converges or the number of iterations reaches a preset number. Among them, the loss function is:
[0047]
[0048] In the formula, L is the cross-entropy loss, t is the number of samples, y (i) is the actual sample label, is the label discriminated by the Softmax layer.
[0049] To further optimize the above technical solution, this embodiment also provides an intelligent detection system for the surface damage of a photovoltaic panel, including: an image acquisition module and a damage detection module. Among them, the image acquisition module is used to collect the surface image of the photovoltaic panel to be detected based on the image acquisition device carried by the drone; the damage detection module is used to input the surface image into the damage detection model and output the surface damage area and damage category of the photovoltaic panel to be detected.
[0050] Furthermore, the damage detection model includes a layer-by-layer feature extraction module, a global feature extraction module, and a damage discrimination module. The layer-by-layer feature extraction module is used to extract damage features at different levels from the surface image; the global feature extraction module is used to obtain the global damage feature based on the damage features at different levels; the damage discrimination module is used to judge the damage category and damaged area based on the global damage feature.
[0051] Among them, the layer-by-layer feature extraction module includes: a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, and a fourth feature extraction layer. Among them, the first feature extraction layer preprocesses the surface image through 7×7 convolution, batch standard normalization, and max pooling operations; the second feature extraction layer, the third feature extraction layer, and the fourth feature extraction layer respectively perform feature extraction on the surface image processed by the first feature extraction layer through different numbers of 3×3 convolutions and 1×1 convolutions, obtain damage feature vectors and damage category vectors with different dimensions and splice them, and output the feature space vector of the damaged area.
[0052] The global feature extraction module performs damage feature extraction at different levels on the surface image, including:
[0053] After normalizing the damage region feature space vector through layer normalization, the attention mechanism is used to learn the damage attention weights of different damage regions;
[0054] Combining the damage region feature space vector and the damage attention weights of different damage regions to obtain the global damage feature.
[0055] The damage discrimination module determines the damage category and damage region based on the global damage feature, including:
[0056] The global damage feature is sequentially input into two fully connected layers and a Softmax classification layer to output the damage category and damage region of the photovoltaic panel to be detected.
[0057] In this embodiment, by collecting normal and damaged images of the photovoltaic panel surface and preprocessing them, the graph convolutional neural network is trained to construct a damage detection model, which can quickly identify the surface damage region and damage category of the photovoltaic panel, is applicable to the situation of multiple faults on the photovoltaic panel surface, can accurately locate the damage region, effectively avoid various background noise interferences caused by coupling, can improve the photovoltaic power generation efficiency in a timely manner, and reduce the losses caused by faults.
[0058] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.
Claims
1. An intelligent detection method for surface damage of a photovoltaic panel, characterized in that, Including: Obtain the surface image of the photovoltaic panel to be detected; Input the surface image into a preset damage detection model, and output the surface damage area and damage category of the photovoltaic panel to be detected; wherein, the damage detection model is obtained by training based on a training set, the training set includes a number of normal pictures of the photovoltaic panel surface and damaged pictures of the photovoltaic panel surface, and the damage detection model is constructed by a graph neural network.
2. The intelligent detection method for surface damage of a photovoltaic panel according to claim 1, wherein Before training the damage detection model based on the training set, it also includes preprocessing the training set. Performing the preprocessing includes: Clean, screen, label the damage area and damage category of the collected normal pictures of the photovoltaic panel surface and damaged pictures of the photovoltaic panel surface, and then perform color correction, random rotation, and random noise addition to complete the preprocessing.
3. The intelligent detection method for surface damage of a photovoltaic panel according to claim 1, wherein The damage detection model includes: a layer-by-layer feature extraction module, a global feature extraction module, and a damage discrimination module. Among them, the layer-by-layer feature extraction module is used to extract damage features at different levels from the surface image; the global feature extraction module is used to obtain global damage features based on damage features at different levels; the damage discrimination module is used to judge the damage category and damage area based on the global damage features.
4. The intelligent detection method for surface damage of a photovoltaic panel according to claim 3, characterized in that, The layer-by-layer feature extraction module includes: a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, and a fourth feature extraction layer. Among them, the first feature extraction layer preprocesses the surface image through 7×7 convolution, batch standard normalization, and max pooling operations; the second feature extraction layer, the third feature extraction layer, and the fourth feature extraction layer respectively perform feature extraction on the surface image processed by the first feature extraction layer through different numbers of 3×3 convolutions and 1×1 convolutions, obtain damage feature vectors and damage category vectors of different dimensions and splice them, and output a damage area feature space vector.
5. The intelligent detection method for surface damage of a photovoltaic panel according to claim 4, wherein, The global feature extraction module extracting damage features at different levels from the surface image includes: After normalizing the damage area feature space vector through layer standard normalization, then using an attention mechanism to learn the damage attention weights of different damage areas; Combining the damage area feature space vector and the damage attention weights of different damage areas to obtain the global damage feature.
6. The intelligent detection method for surface damage of a photovoltaic panel according to claim 4, characterized in that, The damage discrimination module judging the damage category and damage area based on the global damage feature includes: Inputting the global damage feature into two fully connected layers and a Softmax classification layer in sequence, and outputting the damage category and damage area of the photovoltaic panel to be detected.
7. The intelligent detection method for surface damage of a photovoltaic panel according to claim 1, characterized in that, During the process of training the damage detection model, it is judged that the training of the damage detection model is completed by the convergence of the loss function or the iteration reaching a preset number of times. Among them, the loss function is: Where L is the cross-entropy loss, t is the number of samples, and y (i) is the actual sample label, and is the label discriminated by the Softmax layer.
8. An intelligent detection system for surface damage of a photovoltaic panel, which is used to implement the intelligent detection method for surface damage of a photovoltaic panel according to any one of claims 1-7, characterized in that, Including: An image acquisition module and a damage detection module, wherein the image acquisition module is used to acquire the surface image of the photovoltaic panel to be detected based on an image acquisition device carried by a drone; the damage detection module is used to input the surface image into a damage detection model and output the surface damage area and damage category of the photovoltaic panel to be detected; wherein the damage detection model includes a layer-by-layer feature extraction module, a global feature extraction module, and a damage discrimination module, the layer-by-layer feature extraction module is used to extract damage features at different levels from the surface image; the global feature extraction module is used to obtain global damage features based on the damage features at different levels; the damage discrimination module is used to judge the damage category and damage area based on the global damage features.