CN-122023937-A - Intelligent recognition system and method for multi-type retired photovoltaic modules

CN122023937ACN 122023937 ACN122023937 ACN 122023937ACN-122023937-A

Abstract

The invention discloses an intelligent recognition system and method for multi-type retired photovoltaic modules, which belong to the technical field of photovoltaic waste management and solve the problems that the existing method only judges the backboard removal effect through image recognition, the detection object is limited to the local processing quality of a module backboard layer and cannot evaluate and recognize the integral characteristics of the module; according to the invention, the component scanning identification parameters containing the global information of the photovoltaic component are obtained through scanning, multiple complex defects and components of different types are automatically learned and identified through the component identification model, and the component scanning identification parameters are identified and analyzed while the detection precision is improved.

Inventors

YANG YANG
YU BAOYUN
ZHANG YONGXIN
WANG AICHEN
GAO XIANG
LI LIANMING
GAO YAN
ZHENG CHENGHANG
WENG WEIGUO
LI KAI
FENG HONG
MENG ZHIHAO

Assignees

浙江大学
浙江大学嘉兴研究院
浙江物产环保能源股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260228

Claims (10)

1. The intelligent identification method for the multi-type retired photovoltaic modules is characterized by comprising the following steps of: s10, automatically feeding the photovoltaic module, enabling an intelligent identification end to scan the photovoltaic module in real time, acquiring module scanning identification parameters, and uploading the module scanning identification parameters to a database in real time, wherein the intelligent identification end comprises a lower identification module and an upper identification module; S20, pre-constructing a component recognition model based on deep learning, grabbing a modeling sample set from a database, iteratively training the component recognition model by adopting the modeling sample set, and outputting a converged component recognition model; S30, taking the component scanning identification parameters as input, carrying out component scanning identification parameter identification analysis on the component by the component identification model, and outputting a component identification analysis result; S40, carrying out attitude positioning and correction on the photovoltaic module by taking the module identification analysis result as constraint, and triggering a module disassembly instruction based on the attitude positioning and correction result; S50, at least one group of component disassembly modules responds to the component disassembly instruction, executes the photovoltaic component automatic disassembly instruction, and conveys the disassembled photovoltaic component to different processing stations based on the component identification analysis result.
2. The intelligent recognition method of the multi-type retired photovoltaic module according to claim 1, wherein the iterative training method of the module recognition model by using the modeling sample set comprises the following steps: obtaining a modeling sample set, carrying out enhancement treatment on the modeling sample set, and dividing the modeling sample set after the enhancement treatment into a training set and a testing set, wherein the ratio of the training set to the testing set is 3:1; Loading a pre-constructed component identification model, and presetting training rounds, super parameters, an activation function and a joint loss function of the component identification model; Loading a training set, freezing a self-encoder, pre-training a convolutional neural network and a weighted bidirectional feature pyramid network, freezing the convolutional neural network and the weighted bidirectional feature pyramid network after the pre-training, training a type identification layer and a super-resolution reconstruction layer by adopting an alternate optimization strategy, performing full-model joint fine tuning on a component identification model, and outputting a converged component identification model; Obtaining a test set, taking the test set as input, executing the component identification model, outputting a test result, judging whether the test result accords with a preset test precision threshold, outputting a converged component identification model if the test result accords with the preset test precision threshold, and optimizing the super parameters of the component identification model by adopting an Adam optimizer if the test result does not accord with the preset test precision threshold.
3. The intelligent recognition method of the multi-type retired photovoltaic module is characterized in that when the deep learning-based module recognition model is pre-built, a convolutional neural network is taken as a basic framework, the convolutional neural network comprises a convolutional layer, a pooling layer, a SSNBDL network and a full-connection layer which are sequentially connected, the convolutional neural network is taken as an initial model of the module recognition model, the initial model further comprises an input layer and an output layer, a self-encoder is introduced between the input layer and the convolutional neural network, the self-encoder is used for encoding and labeling module scanning recognition parameters, a weighted bidirectional feature pyramid network is introduced between the convolutional neural network and the output layer, the weighted bidirectional feature pyramid network captures the defect area characteristics of the photovoltaic module based on a CBAM attention mechanism, a type recognition layer is introduced between the weighted bidirectional feature pyramid network and the output layer, the type recognition layer is embedded in resnet and csresnext structures, resnet and csresnext structures are respectively used for recognizing the defect types of the photovoltaic module and recognizing the model of the photovoltaic module, a super-resolution reconstruction layer is introduced between the type recognition layer and the output layer, the super-resolution reconstruction layer is based on a deep learning super-resolution virtual convolution module, the cross entropy function is a loss function, the loss function is a two-mean square function loss function, and a two-square loss function is obtained.
4. The intelligent recognition method for the multi-type retired photovoltaic modules according to claim 3, wherein the module recognition model scans the modules to recognize parameters, and the intelligent recognition method comprises the following steps: acquiring component scanning identification parameters, carrying out noise reduction pretreatment on the component scanning identification parameters, and outputting a noise reduction parameter set; The self-encoder performs compression encoding processing on the noise reduction parameter set to generate a low-dimensional feature vector, and adopts a convolutional neural network to perform feature extraction on the low-dimensional feature vector to extract local spatial features of the photovoltaic module, wherein the local spatial features of the photovoltaic module comprise glass morphological features, junction box contour features, frame texture features and defect contour features; generating a component global feature map based on the local spatial features of the photovoltaic component, enhancing the robustness of the local spatial features through SSNBDL networks, and outputting the component global feature map after the scanning noise is restrained; loading a component global feature map, capturing the defect region features of the photovoltaic component based on CBAM attention mechanisms by a weighted bidirectional feature pyramid network, fusing the local space features of the photovoltaic component with multiple scales, and outputting CBAM attention mechanisms to weight and position the component global feature map of the defect region; acquiring CBAM a component global feature map for weighted positioning of a defect region by an attention mechanism, respectively identifying the defect type and the component model of the defect region based on a resnet architecture and a csresnext architecture, and outputting probability distribution of the defect region and the model of the photovoltaic component; And generating a high-resolution virtual component model based on the probability distribution of the defect area and the model of the photovoltaic component, performing secondary detection on the micro defects of the high-resolution virtual component model through a var_threshold function, and outputting a component identification analysis result, wherein the component identification analysis result comprises a component comprehensive evaluation value, a defect area position, a component model, a glass state, a three-dimensional size, a junction box positioning result and a frame positioning result.
5. The intelligent recognition method of the multi-type retired photovoltaic modules according to claim 4, wherein the method for preprocessing the module scanning recognition parameters for noise reduction comprises the following steps: loading the component scanning identification parameters, and processing abnormal values and missing values of the component scanning identification parameters to obtain component scanning identification parameters after the abnormal values and missing values are processed; and carrying out bilateral filtering treatment on the component scanning identification parameters, carrying out contrast increasing treatment on the component scanning identification parameters subjected to bilateral filtering treatment, carrying out filtering treatment on the component scanning identification parameters subjected to contrast increasing treatment through a low-pass filter, and outputting the component scanning identification parameters subjected to filtering treatment.
6. The intelligent recognition method of the multi-type retired photovoltaic modules according to claim 5, wherein the method for preprocessing the module scanning recognition parameters for noise reduction further comprises the following steps: Acquiring a component scanning identification parameter after filtering processing, determining an image gray level threshold based on an image gray level histogram, extracting a pixel gray level value in the component scanning identification parameter, and comparing the pixel gray level value in the component scanning identification parameter with the image gray level threshold to generate a component scanning identification parameter with the image gray level threshold; And adopting a principal component analysis method to perform component scanning identification parameter dimension reduction processing, and outputting component scanning identification parameters after dimension reduction processing.
7. The intelligent recognition method of the multi-type retired photovoltaic module according to claim 6, wherein the method for performing gesture positioning and correction on the photovoltaic module comprises the following steps: Obtaining a component identification analysis result, identifying a defective region position, a component model, a junction box positioning result and a frame positioning result in the component identification analysis result, and extracting a pixel-level edge of a high-resolution virtual component model by adopting a second-order differential algorithm, wherein when the pixel-level edge of the high-resolution virtual component model is extracted, calculating the amplitude and the direction of a gradient by utilizing the second-order differential method, performing non-maximum suppression on the gradient strength, and selecting an edge with the optimal distance from all edge contours of the high-resolution virtual component model combined collineation as the pixel-level edge; loading pixel-level edges of the high-resolution virtual component model, performing edge fitting calculation on the pixel-level edges, and extracting a virtual central axis of the high-resolution virtual component model by filtering interference points of the high-resolution virtual component model based on the base weight; Extracting pixel-level edges and virtual central axes of the high-resolution virtual component model, determining the vertical foot coordinates of the pixel-level edges and the virtual central axes, and affine transforming the vertical foot coordinates of the pixel-level edges and the virtual central axes into pose information of the photovoltaic component; And acquiring pose information of the photovoltaic module, correcting the pose information of the photovoltaic module based on a gray balance algorithm, and outputting the corrected pose information of the photovoltaic module.
8. The intelligent recognition system for the multi-type retired photovoltaic modules is used for implementing the intelligent recognition method for the multi-type retired photovoltaic modules according to any one of claims 1-7, and is characterized by comprising the following components: the intelligent identification end automatically feeds the photovoltaic module, and the intelligent identification end scans the photovoltaic module in real time, acquires module scanning identification parameters and uploads the module scanning identification parameters to the database in real time; the component identification module takes the component scanning identification parameters as input, and the component identification model carries out component scanning identification parameter identification analysis and outputs a component identification analysis result; the gesture positioning module is used for positioning and correcting the gesture of the photovoltaic module by taking the module identification analysis result as constraint and triggering a module disassembly instruction based on the gesture positioning and correction result; And the component disassembly module is used for responding to the component disassembly instruction, executing the photovoltaic component automatic disassembly instruction and conveying the disassembled photovoltaic component to different processing stations based on the component identification analysis result.
9. The intelligent recognition system of the multi-type retired photovoltaic module of claim 8, wherein the intelligent recognition terminal comprises a lower recognition module and an upper recognition module, wherein the lower recognition module is used for scanning the size of the module, recognizing the single-glass and double-glass modules and recognizing the shape of glass, and the upper recognition module is used for scanning the size of the module, recognizing the single-glass and double-glass modules and positioning the junction box and the frame dismantling position.
10. The intelligent recognition system of the multi-type retired photovoltaic module of claim 9, wherein the gesture positioning module comprises: The edge determining unit is used for obtaining a component identification analysis result, identifying the position of a defective area, the model of the component, the positioning of a junction box and the positioning result of a frame in the component identification analysis result, and extracting the pixel-level edge of the high-resolution virtual component model by adopting a second-order differential algorithm; The central axis determining unit is used for loading pixel-level edges of the high-resolution virtual component model, performing edge fitting calculation on the pixel-level edges, filtering interference points of the high-resolution virtual component model based on the graph base weight, and extracting a virtual central axis of the high-resolution virtual component model; The pose extraction unit is used for extracting pixel-level edges and virtual central axes of the high-resolution virtual component model, determining the vertical foot coordinates of the pixel-level edges and the virtual central axes, and affine transforming the vertical foot coordinates of the pixel-level edges and the virtual central axes into pose information of the photovoltaic component; The pose correction unit is used for acquiring pose information of the photovoltaic module, correcting the pose information of the photovoltaic module based on a gray balance algorithm and outputting the corrected pose information of the photovoltaic module.

Description

Intelligent recognition system and method for multi-type retired photovoltaic modules Technical Field The invention belongs to the technical field of photovoltaic waste management, and particularly relates to an intelligent recognition system and method for multiple types of retired photovoltaic modules. Background With rapid development and large-scale application of the photovoltaic industry, photovoltaic modules put into use in early stages are gradually entering the retirement period. It is expected that in the next decades, a huge number of retired photovoltaic modules will be produced worldwide. How to efficiently, environmentally-friendly and economically recycle and treat the retired components to realize the recycling of resources becomes a key problem to be solved urgently in the industry, and has important significance for promoting the development of green low carbon and recycling economy. In the recycling process flow of the photovoltaic module, identification and classification are the first steps of importance. Retired photovoltaic modules are various in variety and mainly comprise two major types of single-glass modules and double-glass modules. These two types of components differ significantly in structure (e.g., number of glass layers, packaging material), disassembly process, and subsequent processing routes. In addition, during transportation and storage of the assembly during service or after retirement, the glass cover plate of the assembly may be in different states such as complete, broken and even crushed, and the frame and the junction box may be damaged or have different positions. These differences and conditions directly affect the selection of subsequent dismantling equipment, the setting of dismantling parameters and the material flow direction. Chinese patent CN119919356A discloses a method for recycling a photovoltaic module and an auxiliary visual detection method, wherein the auxiliary visual detection method comprises the steps of obtaining a first image of the photovoltaic module before removing a back plate layer, identifying the position of a bus bar in the outline of the photovoltaic module according to the first image so as to measure thickness and grind and remove the back plate layer, obtaining a second image of the photovoltaic module after removing the back plate layer, identifying a underworn area or an overworn area of one side of a battery piece, which is opposite to a glass layer, according to the second image, judging the removing effect of the back plate layer according to the underworn area or the overworn area, wherein the conventional method only judges the removing effect of the back plate through image identification, and the detection object is limited to the local processing quality of the back plate layer of the module, so that comprehensive module information cannot be provided for subsequent disassembly due to the fact that the detection object is limited to the local processing quality of the back plate layer (glass state, frame integrity, junction box position, model identification and the like). Disclosure of Invention The invention aims to provide an intelligent recognition system and method for multi-type retired photovoltaic modules, which solve the problems that the existing method only judges the backboard removal effect through image recognition, the detection object is limited to the local processing quality of a module backboard layer, and the whole characteristic of the module cannot be evaluated and recognized, so that comprehensive module information cannot be provided for subsequent disassembly. The intelligent identification method of the multi-type retired photovoltaic modules is realized in such a way that the method comprises the following steps: s10, automatically feeding the photovoltaic module, enabling an intelligent identification end to scan the photovoltaic module in real time, acquiring module scanning identification parameters, and uploading the module scanning identification parameters to a database in real time, wherein the intelligent identification end comprises a lower identification module and an upper identification module; S20, pre-constructing a component recognition model based on deep learning, grabbing a modeling sample set from a database, iteratively training the component recognition model by adopting the modeling sample set, and outputting a converged component recognition model; S30, taking the component scanning identification parameters as input, carrying out component scanning identification parameter identification analysis on the component by the component identification model, and outputting a component identification analysis result; S40, carrying out attitude positioning and correction on the photovoltaic module by taking the module identification analysis result as constraint, and triggering a module disassembly instruction based on the attitude positioning and correction result; S50, at least