Method and system for identifying target components based on a photovoltaic power station

Abstract

The application discloses a kind of based on the identification method and system of target component identification model of photovoltaic power station, the surface image of target component of photovoltaic power station is input to the target component identification model of photovoltaic power station, wherein the target component identification model of photovoltaic power station is based on the surface image of previous target component of photovoltaic power station, type parameter training is formed;Corresponding type parameter is output based on the target component identification model of photovoltaic power station;According to type parameter and type matching table definition photovoltaic power station target component type, wherein the target component type of photovoltaic power station includes photovoltaic component and non-photovoltaic component, photovoltaic component includes photovoltaic board, briquetting or mounting support, at this time, the surface image of target component of photovoltaic power station is intelligently identified based on the target component identification model of photovoltaic power station, according to type parameter and type matching table definition photovoltaic power station target component type, improve the identification accuracy of target component type of photovoltaic power station.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of identification of photovoltaic cleaning robots, and particularly relates to an identification method and system based on a target component identification model of a photovoltaic power station. BACKGROUND

[0002] With the development of science and technology, photovoltaic cleaning robots travel on the surface of photovoltaic panels and clean the surface of photovoltaic panels. The photovoltaic cleaning robot is provided with a camera, and an image is collected through the camera. At this time, the collected image can contain photovoltaic components and non-photovoltaic components. In the prior art, the outer contour is highlighted for the image, and matching is performed according to the outer contour to define the type of photovoltaic components or non-photovoltaic components. However, there are various situations of target components of the photovoltaic power station with similar outer contours in the image, and single contour identification will affect the identification accuracy of the type of target components of the photovoltaic power station. SUMMARY

[0003] The present application aims to overcome the shortcomings of the prior art, and provides an identification method and system based on a target component identification model of a photovoltaic power station. The target component identification model of the photovoltaic power station intelligently identifies the surface image of the target component of the photovoltaic power station and outputs the corresponding type parameter, so as to define the type of the target component of the photovoltaic power station according to the type parameter and the type matching table, thereby clearly defining the type of the target component of the photovoltaic power station, realizing intelligent identification of the surface image of the target component of the photovoltaic power station, and improving the identification accuracy of the type of the target component of the photovoltaic power station.

[0004] To solve the above technical problems, the present application provides an identification method based on a target component identification model of a photovoltaic power station, which is applied to the scenario of identifying the target component of the photovoltaic power station by a photovoltaic cleaning robot. The identification method based on the target component identification model of the photovoltaic power station comprises the following steps:

[0005] obtaining a surface image of a target component of a photovoltaic power station;

[0006] inputting the surface image of the target component of the photovoltaic power station into a target component identification model of the photovoltaic power station, wherein the target component identification model of the photovoltaic power station is trained based on the surface image and type parameter of the target component of the photovoltaic power station in the past;

[0007] outputting a corresponding type parameter based on the target component identification model of the photovoltaic power station;

[0008] defining the type of the target component of the photovoltaic power station according to the type parameter and a type matching table, wherein the type of the target component of the photovoltaic power station comprises photovoltaic components and non-photovoltaic components, and the photovoltaic components comprise photovoltaic panels, pressing blocks or mounting supports.

[0009] Optionally, the surface image of the target component of the photovoltaic power station is acquired, comprising:

[0010] The camera of the photovoltaic cleaning robot is triggered when the photovoltaic cleaning robot moves relative to the photovoltaic panel;

[0011] The camera is externally photographed when the photovoltaic cleaning robot moves relative to the photovoltaic panel to acquire the surface image of the target component of the photovoltaic power station.

[0012] Optionally, the surface image of the target component of the photovoltaic power station is input to a target component identification model of the photovoltaic power station, wherein the target component identification model of the photovoltaic power station is trained based on the surface image of the target component of the photovoltaic power station in the past and the type parameter, and comprises:

[0013] The surface image of the target component of the photovoltaic power station is input to a target component identification model of the photovoltaic power station, wherein the target component identification model of the photovoltaic power station is trained based on the surface image of the target component of the photovoltaic power station in the past and the type parameter, and comprises:

[0014] The surface image of the target component of the photovoltaic power station is input to a target component identification model of the photovoltaic power station, wherein the target component identification model of the photovoltaic power station is trained based on the surface image of the target component of the photovoltaic power station in the past and the type parameter.

[0015] Optionally, the surface image of the target component of the photovoltaic power station is input to a target component identification model of the photovoltaic power station, wherein the target component identification model of the photovoltaic power station is trained based on the surface image of the target component of the photovoltaic power station in the past and the type parameter, and further comprises:

[0016] For training of the target component identification model of the photovoltaic power station, the surface images of the target components of a plurality of photovoltaic power stations in the past are collected;

[0017] An image data set is formed based on the surface images of the target components of the plurality of photovoltaic power stations in the past;

[0018] The image data set is labeled to determine the type parameter of each surface image;

[0019] The image data set is divided into a training set, a validation set and a test set, wherein the training set, the validation set and the test set have a certain ratio;

[0020] A known target component identification model is collected;

[0021] The known target component identification model learns the model parameters in the training set to complete the training of the known target component identification model in the current round in the training set;

[0022] After the known target component recognition model completes the training on the training set in the current round, the known target component recognition model performs model verification and iteration of training hyperparameters in the verification set, the training hyperparameters including a learning rate, network width and depth, and various training weight parameters;

[0023] After the known target component recognition model completes the training and all hyperparameter optimization, the known target component recognition model is tested for generalization ability based on the test set to estimate a generalization error value of the known target component recognition model,

[0024] If the generalization error value is within a preset generalization error threshold, the target component recognition model of the photovoltaic power station is output.

[0025] Optionally, the target component recognition model of the photovoltaic power station outputs corresponding type parameters, including:

[0026] Based on the surface image of the target component of the photovoltaic power station, a plurality of candidate bounding boxes are divided, wherein the candidate bounding boxes are labeled with corresponding type parameters;

[0027] The plurality of candidate bounding boxes are subjected to bounding box confidence filtering to determine test candidate bounding boxes;

[0028] Distribution focus loss calculation is performed according to the test candidate bounding boxes to determine focus loss values of the test candidate bounding boxes;

[0029] If the focus loss value of the test candidate bounding box is less than a preset focus loss threshold, the test candidate bounding box is determined as a standard candidate bounding box, and the type parameter corresponding to the standard candidate bounding box is determined.

[0030] Optionally, the target component type of the photovoltaic power station is defined according to the type parameter and a type matching table, wherein the target component type of the photovoltaic power station includes a photovoltaic component and a non-photovoltaic component, and the photovoltaic component includes a photovoltaic panel, a pressing block, or a mounting bracket, including:

[0031] The type parameter corresponding to the standard candidate bounding box is collected;

[0032] The type parameter corresponding to the standard candidate bounding box is associated with the type matching table;

[0033] The target component type of the photovoltaic power station is determined according to the type parameter corresponding to the standard candidate bounding box and the type matching table.

[0034] Optionally, the target component type of the photovoltaic power station is defined according to the type parameter and a type matching table, wherein the target component type of the photovoltaic power station includes a photovoltaic component and a non-photovoltaic component, and the photovoltaic component includes a photovoltaic panel, a pressing block, or a mounting bracket, and further including:

[0035] If the target component type of the photovoltaic power station is a photovoltaic panel, the intersection over union of the standard candidate bounding box and the upper half of the surface image of the target component of the photovoltaic power station is calculated;

[0036] If the intersection over union is 0, there is no photovoltaic component on the surface image of the target component of the photovoltaic power station, and the anti-falling strategy of the photovoltaic cleaning robot is triggered;

[0037] If the intersection over union is greater than 0, there is a photovoltaic component on the surface image of the target component of the photovoltaic power station, and the photovoltaic cleaning robot continues to walk.

[0038] Optionally, the target component type of the photovoltaic power station is defined according to the type parameter and the type matching table, wherein the target component type of the photovoltaic power station includes a photovoltaic component and a non-photovoltaic component, the photovoltaic component includes a photovoltaic panel, a pressing block or a mounting bracket, and the method further comprises:

[0039] The calculation formula of the intersection over union of the standard candidate bounding box and the upper half of the surface image of the target component of the photovoltaic power station is as follows:

[0040] h intersection =max[0,min(Ay2,By2)-max(Ay1,By1)+1]

[0041] w intersection =max[0,min(Ax2,Bx2)-max(Ax1,Bx1)+1]

[0042] I=h intersection ×w intersection

[0043] U=(|Ay2-Ay1|+1)×(|Ax2-Ax1|+1)+(|By2-By1|+1)×(|Bx2-Bx1|+1)-I

[0044]

[0045] wherein,

[0046] h intersection represents the height of the standard candidate bounding box; w intersection represents the width of the standard candidate bounding box; I represents the area of the intersection; max[0,min(Ax2,Bx2) represents taking the larger one, min(Ay2,By2) represents taking the smaller one, |Bx2-Bx1| represents the corresponding absolute value, and IoU is the intersection over union.

[0047] Optionally, the identification method based on the target component identification model of the photovoltaic power station further comprises:

[0048] inputting the surface image of the target component of the photovoltaic power station to a target component identification model of the photovoltaic power station;

[0049] If the target component identification model of the photovoltaic power station outputs an unrecognizable signal, triggering an anti-falling strategy of the photovoltaic cleaning robot according to the unrecognizable signal.

[0050] Optionally, a recognition system based on a target component identification model of a photovoltaic power station, characterized in that the recognition system based on the target component identification model of the photovoltaic power station is applied to the recognition method based on the target component identification model of the photovoltaic power station, and the recognition system based on the target component identification model of the photovoltaic power station comprises:

[0051] The acquisition module is configured to acquire the surface image of the target component of the photovoltaic power station.

[0052] The input module is configured to input the surface image of the target component of the photovoltaic power station to a target component identification model of the photovoltaic power station, wherein the target component identification model of the photovoltaic power station is trained based on the surface image of the target component of the photovoltaic power station and type parameters in the past.

[0053] The output module is configured to output corresponding type parameters based on the target component identification model of the photovoltaic power station.

[0054] The target component type module is configured to define the type of the target component of the photovoltaic power station according to the type parameters and a type matching table, wherein the type of the target component of the photovoltaic power station comprises a photovoltaic component and a non-photovoltaic component, and the photovoltaic component comprises a photovoltaic panel, a pressing block or a mounting bracket.

[0055] In the embodiment of the application, the surface image of the target component of the photovoltaic power station is acquired, the surface image of the target component of the photovoltaic power station is inputted to the target component identification model of the photovoltaic power station, the corresponding type parameters are outputted based on the target component identification model of the photovoltaic power station, and the type of the target component of the photovoltaic power station is defined according to the type parameters and the type matching table, wherein the type of the target component of the photovoltaic power station comprises a photovoltaic component and a non-photovoltaic component, and the photovoltaic component comprises a photovoltaic panel, a pressing block or a mounting bracket. At this time, the surface image of the target component of the photovoltaic power station is intelligently recognized based on the target component identification model of the photovoltaic power station, and the corresponding type parameters are outputted, so as to define the type of the target component of the photovoltaic power station according to the type parameters and the type matching table, thereby clearly defining the type of the target component of the photovoltaic power station, realizing the intelligent recognition of the surface image of the target component of the photovoltaic power station, and improving the recognition accuracy of the type of the target component of the photovoltaic power station. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced. Obviously, the accompanying drawings in the following description only only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the premise of the accompanying drawings.

[0057] Figure 1 is a flowchart of the identification method of the target component identification model based on the photovoltaic power station in the embodiments of the present application;

[0058] Figure 2 is a flowchart of S11 in the identification method of the target component identification model based on the photovoltaic power station in the embodiments of the present application;

[0059] Figure 3 is a flowchart of S12 in the identification method of the target component identification model based on the photovoltaic power station in the embodiments of the present application;

[0060] Figure 4 is a flowchart of S13 in the identification method of the target component identification model based on the photovoltaic power station in the embodiments of the present application;

[0061] Figure 5 is a flowchart of S14 in the identification method of the target component identification model based on the photovoltaic power station in the embodiments of the present application;

[0062] Figure 6 is a structural composition diagram of the identification system of the target component identification model based on the photovoltaic power station in the embodiments of the present application;

[0063] Figure 7 is a hardware diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0064] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0065] EMBODIMENT

[0066] Please refer to Figures 1 to 7 A target component identification model identification method based on a photovoltaic power station is applied to a scenario in which a photovoltaic cleaning robot identifies target components of a photovoltaic power station. The target component identification model identification method based on the photovoltaic power station comprises:

[0067] Step S11: obtaining a surface image of a target component of a photovoltaic power station;

[0068] Step S12: inputting the surface image of the target component of the photovoltaic power station into a target component recognition model of the photovoltaic power station, wherein the target component recognition model of the photovoltaic power station is trained based on a surface image of a previous target component of the photovoltaic power station and a type parameter;

[0069] Step S13: outputting a corresponding type parameter based on the target component recognition model of the photovoltaic power station;

[0070] Step S14: defining a target component type of the photovoltaic power station according to the type parameter and a type matching table, wherein the target component type of the photovoltaic power station includes a photovoltaic component and a non-photovoltaic component, and the photovoltaic component includes a photovoltaic panel, a pressing block or a mounting bracket.

[0071] In the embodiment of the present application, the surface image of the target component of the photovoltaic power station is obtained, the surface image of the target component of the photovoltaic power station is inputted into the target component recognition model of the photovoltaic power station, wherein the target component recognition model of the photovoltaic power station is trained based on a surface image of a previous target component of the photovoltaic power station and a type parameter, a corresponding type parameter is outputted based on the target component recognition model of the photovoltaic power station, and a target component type of the photovoltaic power station is defined according to the type parameter and a type matching table, wherein the target component type of the photovoltaic power station includes a photovoltaic component and a non-photovoltaic component, and the photovoltaic component includes a photovoltaic panel, a pressing block or a mounting bracket. At this time, the surface image of the target component of the photovoltaic power station is intelligently recognized based on the target component recognition model of the photovoltaic power station, and a corresponding type parameter is outputted, so as to define the target component type of the photovoltaic power station according to the type parameter and the type matching table, thereby clearly defining the target component type of the photovoltaic power station, and realizing intelligent recognition of the surface image of the target component of the photovoltaic power station and improving the recognition accuracy of the target component type of the photovoltaic power station.

[0072] Please refer to Figure 2 In step S11, the surface image of the target component of the photovoltaic power station is obtained.

[0073] In the specific implementation process of the present application, the specific steps can be:

[0074] S111: triggering a camera of a photovoltaic cleaning robot when the photovoltaic cleaning robot moves relative to a photovoltaic panel;

[0075] S112: taking pictures outside by the camera when the photovoltaic cleaning robot moves relative to the photovoltaic panel, so as to obtain the surface image of the target component of the photovoltaic power station.

[0076] In the embodiment of the present application, the photovoltaic cleaning robot drives in the photovoltaic power station and moves on the surface of the plurality of photovoltaic panels along the preset path, at this time, the photovoltaic cleaning robot passes through the plurality of photovoltaic panels and the connecting part between the adjacent two photovoltaic panels in turn.

[0077] When the photovoltaic cleaning robot moves relative to the photovoltaic panel, the camera of the photovoltaic cleaning robot is triggered, the camera is in the starting state, and the camera can be arranged at the bottom of the photovoltaic cleaning robot or the side of the photovoltaic cleaning robot, which is not limited here.

[0078] At this time, the camera shoots outward when the photovoltaic cleaning robot moves relative to the photovoltaic panel to obtain the surface image of the target component of the photovoltaic power station, so as to facilitate subsequent processing of the surface image of the target component of the photovoltaic power station, so as to take the surface image of the target component of the photovoltaic power station as an input image, so as to intelligently identify the surface image of the target component of the photovoltaic power station.

[0079] Optionally, when the photovoltaic cleaning robot moves relative to the photovoltaic panel, a plurality of pose parameters of the photovoltaic cleaning robot are collected; the attitude of the photovoltaic cleaning robot in the moving process is determined according to the plurality of pose parameters; the inclination angle of the photovoltaic panel is determined based on the attitude of the photovoltaic cleaning robot in the moving process; when the camera end of the camera is directly opposite the target component of the photovoltaic power station, the camera is triggered to shoot the photovoltaic power station to determine the surface image of the photovoltaic power station.

[0080] Please refer to Figure 3 In step S12, the surface image of the target component of the photovoltaic power station is input to the target component identification model of the photovoltaic power station, wherein the target component identification model of the photovoltaic power station is trained based on the surface image of the target component of the photovoltaic power station and the type parameter in the past;

[0081] In the specific implementation process of the present application, the specific steps can be:

[0082] S121: determining the target component preprocessing image of the photovoltaic power station based on the surface image of the photovoltaic power station, wherein the collected surface image of the target component of the photovoltaic power station is preprocessed, and overexposed images, underexposed images and blurred images are deleted;

[0083] S122: inputting the target component preprocessing image of the photovoltaic power station to the target component identification model of the photovoltaic power station, and identifying the component type by the target component identification model of the photovoltaic power station, wherein the target component identification model of the photovoltaic power station is trained based on the surface image of the target component of the photovoltaic power station and the type parameter in the past.

[0084] In the embodiment of the present application, the surface image of the photovoltaic power station as the input image, the surface image of the photovoltaic power station can be preprocessed, so as to optimize the surface image of the photovoltaic power station in the preprocessing process, at this time, the surface image of the target component of the photovoltaic power station collected is preprocessed, and the overexposed image, underexposed image and blurred image are deleted, wherein the overexposed image, underexposed image and blurred image are affected images and are deleted to avoid the influence of the overexposed image, underexposed image and blurred image, thereby ensuring the recognition accuracy of the intelligent recognition of the surface image of the photovoltaic power station.

[0085] Further, the target component preprocessing image of the photovoltaic power station is input to the target component recognition model of the photovoltaic power station, and the component type is recognized by the target component recognition model of the photovoltaic power station, wherein the target component recognition model of the photovoltaic power station is trained based on the surface image of the target component of the photovoltaic power station and the type parameter in the past, at this time, the target component preprocessing image of the photovoltaic power station is input to the target component recognition model of the photovoltaic power station as the input image, and the target component recognition model of the photovoltaic power station intelligently recognizes the target component of the photovoltaic power station and outputs the type parameter.

[0086] At this time, the target component recognition model of the photovoltaic power station is a learning model, and intelligently recognizes the surface image of the target component of the photovoltaic power station, and the training of the target component recognition model of the photovoltaic power station is described as follows:

[0087] The first step is to collect the surface images of the target components of a plurality of photovoltaic power stations in the past; form an image data set based on the surface images of the target components of a plurality of photovoltaic power stations in the past; and label the image data set to determine the type parameters of each surface image.

[0088] At this time, the target component recognition model of the photovoltaic power station is trained based on the surface images of the target components of a plurality of photovoltaic power stations in the past, and the surface images of the target components of a plurality of photovoltaic power stations in the past are used as the image data set, so as to further train based on the image data set, thereby ensuring the training accuracy of the target component recognition model of the photovoltaic power station.

[0089] The second step is to divide the image data set into a training set, a validation set and a test set, wherein the training set, the validation set and the test set have a certain proportion.

[0090] The image data set is divided so as to use the training set, the validation set and the test set for subsequent processing, at this time, the training set, the validation set and the test set have a certain proportion, which can be adjusted according to the training data of the target component recognition model of the photovoltaic power station, or can be adjusted according to the actual training scene of the target component recognition model of the photovoltaic power station, which is not limited here.

[0091] At this point, the training set, validation set, and test set have a certain ratio, and the ratio between the training set, validation set, and test set is reasonably divided so as to form a reasonable ratio based on the training set, validation set, and test set. This allows for control over the target component identification model of the photovoltaic power station during the training, validation, and testing phases, thereby ensuring the accuracy of the target component identification model of the photovoltaic power station.

[0092] The third step: Collect known target component recognition models; learn the model parameters of the known target component recognition models in the training set to complete the training of the known target component recognition models in the current round on the training set;

[0093] At this point, a known target component recognition model is introduced, and the known target component recognition model is specifically trained to facilitate in-depth optimization of the known target component recognition model. This ensures the conversion between the known target component recognition model and our target component recognition model, and enables it to be deeply adapted to our application environment.

[0094] Fourth step: After the known target component recognition model completes the training of the training set in the current round, the known target component recognition model performs model validation and iterative training of hyperparameters on the validation set. Hyperparameters include learning rate, network width and depth, and various training weight parameters.

[0095] After training and optimizing all hyperparameters of the known target component recognition model, a generalization ability test is performed on the test set to estimate the generalization error of the known target component recognition model.

[0096] If the generalization error value is within the preset generalization error threshold, the identification model of the target component of the photovoltaic power station will be output.

[0097] At this point, the known target component recognition model is trained on the training set, validated on the validation set, and tested on the test set in sequence, thereby gaining in-depth control over the training, validation, and testing phases.

[0098] Furthermore, the model is specifically trained on known target component recognition models. After each round of training, the model is validated and the training hyperparameters are iterated on the validation set to avoid overfitting the model on the training set and further improve the model's generalization ability.

[0099] Meanwhile, a generalization error value is introduced. After the known target component recognition model has completed training and all hyperparameter optimization, the generalization ability of the known target component recognition model is tested based on the test set to estimate the generalization error value of the known target component recognition model. If the generalization error value is within the preset generalization error threshold, the recognition model of the target component of the photovoltaic power station is output, thereby realizing the output of the recognition model of the target component of the photovoltaic power station and ensuring the accuracy of the recognition model of the target component of the photovoltaic power station.

[0100] Please see Figure 4 In step S13, the corresponding type parameters are output based on the target component identification model of the photovoltaic power station;

[0101] In the specific implementation of this invention, the specific steps can be as follows:

[0102] S131: Divide multiple candidate bounding boxes based on the surface image of the target component of the photovoltaic power station, wherein the candidate bounding boxes have been labeled with the corresponding type parameters;

[0103] S132: Filter multiple candidate bounding boxes using bounding box confidence scores to determine the candidate bounding boxes to be tested;

[0104] S133: Calculate the distributed focus loss based on the candidate bounding boxes to be tested, so as to determine the focus loss value of each candidate bounding box to be tested;

[0105] S134: If the focus loss value of the candidate bounding box to be tested is less than the preset focus loss threshold, then the candidate bounding box to be tested is determined as the standard candidate bounding box, and the type parameter corresponding to the standard candidate bounding box is determined.

[0106] In the embodiments of this application, the surface image of the target component of the photovoltaic power station is divided into multiple candidate bounding boxes to form multiple candidate bounding boxes, and then the multiple candidate bounding boxes are processed in a targeted manner. At this time, the candidate bounding boxes are marked with the corresponding type parameters.

[0107] Furthermore, multiple candidate bounding boxes are filtered using bounding box confidence to determine the candidate bounding boxes to be tested. This filtering of multiple candidate bounding boxes is achieved by using bounding box confidence to filter multiple candidate bounding boxes under the control of bounding box confidence, thereby eliminating some candidate bounding boxes with low confidence and outputting the candidate bounding boxes to be tested, thus realizing the first filtering of bounding boxes.

[0108] Furthermore, the focus loss is calculated based on the candidate bounding boxes to be tested to determine the focus loss value of each candidate bounding box; if the focus loss value of a candidate bounding box to be tested is less than a preset focus loss threshold, then the candidate bounding box to be tested is determined as a standard candidate bounding box, and the type parameter corresponding to the standard candidate bounding box is determined.

[0109] At this point, a distributed focus loss is introduced, and the distributed focus loss is calculated for the candidate bounding boxes to be tested in order to determine the focus loss value of each candidate bounding box. Then, a focus loss value is introduced, and the focus loss value is compared with a preset focus loss threshold to determine the standard candidate bounding box. This realizes the second filtering of the bounding box, thus completing the multiple filtering of the bounding box and ensuring the accuracy of the standard candidate bounding box.

[0110] At this point, multiple candidate bounding boxes are sequentially screened to eliminate unqualified candidate bounding boxes and retain the candidate bounding boxes to be tested. At the same time, standard candidate bounding boxes are determined based on the candidate bounding boxes to be tested and the focus loss, thus realizing the final determination of standard candidate bounding boxes. This facilitates type identification based on the standard candidate bounding boxes and the target component identification model of the photovoltaic power station, thereby ensuring the identification quantity of the target component identification model of the photovoltaic power station, avoiding excessive noise, and demonstrating the accurate identification of the target component identification model of the photovoltaic power station by the standard candidate bounding boxes.

[0111] Furthermore, a distributed focal loss function (DFL) is introduced, moving its structure outside the target component identification model of the photovoltaic power plant. If the DFL structure is placed before "bounding box confidence filtering," the distributed focal loss is calculated for each candidate bounding box. However, if this structure is placed after "bounding box confidence filtering," assuming that only n candidate bounding boxes remain after filtering, the distributed focal loss only needs to be calculated for these n candidate bounding boxes, thus significantly reducing the consumption of computational and bandwidth resources. Since the performance of this calculation on the NPU is generally poor, the distributed focal loss calculation is performed on the CPU, thereby greatly improving the utilization efficiency of the NPU.

[0112] Another method involves retrieving bounding boxes that meet the requirements based on the quantized confidence scores. If the confidence score of a candidate bounding box is less than a preset threshold, it is eliminated. The coordinates of the qualified bounding boxes are then obtained using distributed focus loss. The confidence score of these bounding boxes is dequantized, and the confidence score and classification result are recorded. Bounding boxes obtained under different time lengths are summed and quickly sorted according to their confidence scores. The Non-Maximum Suppression (NMS) algorithm is used to select the bounding boxes with higher confidence scores from those with high overlap regions, and these are output as the final result. The selection criterion is a pre-set NMS threshold. If the LoU of two bounding boxes is greater than the NMS threshold, the one with the higher confidence score is selected.

[0113] Meanwhile, a summation operation for the classified detection targets has been added to the feedforward part of the original YOLOv8 model's head. This is because, assuming there are m candidate bounding boxes and n classes, the threshold retrieval operation would require m*n times. However, by summing the confidence scores for each class, unacceptable confidence scores can be quickly filtered out, significantly reducing the time required for threshold detection.

[0114] At this point, referencing the structure of YOLOv5, the n-class output was adjusted to n+1 classes. The newly added class is used to store the confidence scores of candidate bounding boxes, which can be used to quickly filter the confidence scores of candidate bounding boxes. In this way, during post-processing, when thresholding the confidence scores on the CPU, the number of logical judgments can be reduced by tens of times.

[0115] While activation functions can retain more gradient information and prevent the "vanishing gradient" phenomenon, they require higher computational complexity and perform poorly on NPUs. On the other hand, due to the linearity of the ReLU activation function, quantized models have faster inference speeds on NPUs, providing higher FPS and meeting the real-time requirements of onboard deployments.

[0116] Please see Figure 5 S14: Define the target component type of the photovoltaic power station according to the type parameter and the type matching table. The target component type of the photovoltaic power station includes photovoltaic components and non-photovoltaic components. Photovoltaic components include photovoltaic panels, briquettes or mounting brackets.

[0117] In the specific implementation of this invention, the specific steps can be as follows:

[0118] S141: Collect the type parameters corresponding to the standard candidate bounding boxes;

[0119] S142: Associate the type parameters corresponding to the standard candidate bounding boxes with the type matching table;

[0120] S143: Determine the target component type of the photovoltaic power plant based on the type parameters and type matching table corresponding to the standard candidate bounding box.

[0121] In the embodiments of this application, standard candidate bounding boxes are collected and identified. At this time, type parameters corresponding to the standard candidate bounding box markers are collected to associate the type parameters corresponding to the standard candidate bounding boxes with a type matching table. Selection is made based on the mapping relationship between the type parameters and the type matching table, thereby determining the target component type of the photovoltaic power station according to the type parameters corresponding to the standard candidate bounding boxes and the type matching table. Here, photovoltaic components include photovoltaic panels, briquettes, or mounting brackets, etc., which are not limited here. Non-photovoltaic components include connecting bridges, photovoltaic scenes, maintenance channels, etc., which are not limited here.

[0122] At this point, the surface image of the target component of the photovoltaic power station is intelligently identified based on the target component identification model of the photovoltaic power station, and the corresponding type parameters are output. This allows the target component type of the photovoltaic power station to be defined according to the type parameters and the type matching table, thereby clarifying the target component type of the photovoltaic power station. This achieves intelligent identification of the surface image of the target component of the photovoltaic power station and improves the accuracy of the identification of the target component type of the photovoltaic power station.

[0123] Furthermore, if the target component type of the photovoltaic power station is a photovoltaic panel, the intersection-union ratio (IUU) of the standard candidate bounding box and the upper half of the surface image of the target component of the photovoltaic power station is calculated. At this time, the IUU of the standard candidate bounding box and the upper half of the surface image of the target component of the photovoltaic power station is introduced to define the initial positioning of the photovoltaic component and to focus on the position of the photovoltaic component in order to trigger the corresponding anti-fall strategy. This enables the photovoltaic cleaning robot to perform the corresponding movement logic when identifying the target component, thus ensuring the intelligent movement of the photovoltaic cleaning robot.

[0124] At this time, if the crossover ratio is 0, there are no photovoltaic modules on the surface image of the target module of the photovoltaic power station, and the anti-fall strategy of the photovoltaic cleaning robot is triggered. The anti-fall strategy includes the photovoltaic cleaning robot stopping and moving in the opposite direction to avoid the photovoltaic cleaning robot falling.

[0125] In addition, if the crossover ratio is greater than 0, then there are photovoltaic modules on the surface image of the target module of the photovoltaic power station, and the photovoltaic cleaning robot continues to move.

[0126] Specifically, the formula for calculating the intersection-union ratio (IUR) of the standard candidate bounding box and the upper half of the surface image of the target module of the photovoltaic power plant is as follows:

[0127] h intersection =max[0,min(Ay2,By2)-max(Ay1,By1)+1]

[0128] w intersection =max[0,min(Ax2,Bx2)-max(Ax1,Bx1)+1]

[0129] I = h intersection ×w intersection

[0130] U=(|Ay2-Ay1|+1)×(|Ax2-Ax1|+1)+(|By2-By1|+1)×(|Bx2-Bx1|+1)-I

[0131]

[0132] in,

[0133] h intersection Indicates the height of the standard candidate bounding box; w intersection The width of the standard candidate bounding box is represented by I; the area of ​​the intersection is represented by max[0,min(Ax2,Bx2)], which means taking the larger of the two, min(Ay2,By2) means taking the smaller of the two, |Bx2-Bx1| represents the corresponding absolute value, and IoU is the intersection-union ratio.

[0134] In addition, the identification method based on the target component identification model of the photovoltaic power station also includes: inputting the surface image of the target component of the photovoltaic power station into the target component identification model of the photovoltaic power station; if the target component identification model of the photovoltaic power station outputs an unidentifiable signal, then triggering the anti-fall strategy of the photovoltaic cleaning robot according to the unidentifiable signal.

[0135] At this point, for the surface image of the target component of the photovoltaic power station that cannot be identified, the anti-fall strategy of the photovoltaic cleaning robot is triggered based on the unidentifiable signal, thereby controlling the photovoltaic cleaning robot to drive on the unknown part and ensuring the driving safety performance of the photovoltaic cleaning robot.

[0136] In this embodiment of the invention, the surface image of a target component of a photovoltaic power station is obtained using the method described herein. This surface image is then input into a target component recognition model for the photovoltaic power station. The target component recognition model is trained based on previous surface images and type parameters of target components from photovoltaic power stations. The model outputs corresponding type parameters. The target component type of the photovoltaic power station is defined according to the type parameters and a type matching table. The target component type includes photovoltaic components and non-photovoltaic components. Photovoltaic components include photovoltaic panels, blocks, or mounting brackets. In this process, the target component recognition model intelligently identifies the surface image of the target component and outputs corresponding type parameters. This allows for the definition of the target component type based on the type parameters and the type matching table, thereby clarifying the target component type and achieving intelligent recognition of the surface image of the target component, thus improving the accuracy of target component type recognition.

[0137] Example

[0138] Please see Figure 6 , Figure 6 This is a schematic diagram of the structural composition of the identification system based on the target component identification model of a photovoltaic power station in an embodiment of the present invention.

[0139] like Figure 6 As shown, an identification system based on a target component identification model of a photovoltaic power station is provided. The identification system includes:

[0140] Acquisition module 21 is used to acquire surface images of target components in a photovoltaic power station;

[0141] The input module 22 is used to input the surface image of the target component of the photovoltaic power station into the target component recognition model of the photovoltaic power station. The target component recognition model of the photovoltaic power station is trained based on the surface images and type parameters of the target components of previous photovoltaic power stations.

[0142] Output module 23 is used to output the corresponding type parameters based on the target component identification model of the photovoltaic power station;

[0143] The target component type module 24 is used to define the target component type of the photovoltaic power station according to the type parameter and the type matching table. The target component type of the photovoltaic power station includes photovoltaic components and non-photovoltaic components. Photovoltaic components include photovoltaic panels, briquettes or mounting brackets.

[0144] Example

[0145] Please see Figure 7 See below for reference. Figure 7To describe an electronic device 40 according to this embodiment of the present invention. Figure 7 The electronic device 40 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0146] like Figure 7 As shown, the electronic device 40 is manifested in the form of a general-purpose computing device. The components of the electronic device 40 may include, but are not limited to: at least one processing unit 41, at least one storage unit 42, and a bus 43 connecting different system components (including storage unit 42 and processing unit 41).

[0147] The storage unit stores program code that can be executed by the processing unit 41, causing the processing unit 41 to perform the steps described in the "Embodiment Methods" section of this specification according to various exemplary embodiments of the present invention.

[0148] Storage unit 42 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 421 and / or cache memory 422, and may further include a read-only memory (ROM) 423.

[0149] Storage unit 42 may also include a program / utility 424 having a set (at least one) of program modules 425, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0150] Bus 43 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the multiple bus structures.

[0151] Electronic device 40 can also communicate with one or more external devices (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 40, and / or with any device that enables electronic device 40 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed through input / output (I / O) interface 44. Furthermore, electronic device 40 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) through network adapter 45. Figure 7 As shown, network adapter 45 communicates with other modules of electronic device 40 via bus 43. It should be understood that, although... Figure 7As not shown, other hardware and / or software modules may be used in conjunction with electronic device 40, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup planning systems.

[0152] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the method according to the embodiments of this disclosure.

[0153] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk, etc. Furthermore, it stores computer program instructions, which, when executed by a computer, cause the computer to perform the methods described above.

[0154] Furthermore, the above description provides a detailed introduction to the identification method and system based on the target component identification model of a photovoltaic power station provided in the embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for identifying target components based on a photovoltaic power plant target component identification model, characterized in that, This method is applied to scenarios where photovoltaic cleaning robots identify target components in photovoltaic power plants; the identification method based on the target component identification model of photovoltaic power plants includes: Acquire surface images of target components in a photovoltaic power plant; The surface image of the target component of the photovoltaic power station is input into the target component recognition model of the photovoltaic power station. The target component recognition model of the photovoltaic power station is trained based on the surface images and type parameters of the target components of previous photovoltaic power stations. The target component identification model for photovoltaic power plants outputs corresponding type parameters, including: dividing multiple candidate bounding boxes based on the surface image of the target component of the photovoltaic power plant, wherein the candidate bounding boxes are labeled with corresponding type parameters; performing bounding box confidence filtering on multiple candidate bounding boxes to determine the candidate bounding boxes to be tested; calculating the distributed focus loss based on the candidate bounding boxes to be tested to determine the focus loss value of each candidate bounding box to be tested; if the focus loss value of the candidate bounding box to be tested is less than a preset focus loss threshold, then the candidate bounding box to be tested is determined as a standard candidate bounding box, and the type parameter corresponding to the standard candidate bounding box is determined. The coordinates of the standard bounding boxes are obtained through distributed focus loss. The confidence of the bounding boxes is dequantized, and the confidence and classification results are recorded. The bounding boxes obtained under different time lengths are summed, and these bounding boxes are quickly sorted according to their confidence. The non-maximum suppression algorithm is used to select the bounding boxes with high confidence in the overlapping area that have a confidence greater than the threshold as the result output. The structure of the distributed focus loss function is moved outside the target component recognition model of the photovoltaic power station. The distributed focus loss is calculated on the CPU. The target component type of the photovoltaic power station is defined according to type parameters and a type matching table. The target component type includes photovoltaic (PV) components and non-PV components. PV components include PV panels, mounting blocks, or mounting brackets. The process includes: collecting type parameters corresponding to standard candidate bounding boxes; associating the type parameters corresponding to the standard candidate bounding boxes with a type matching table; determining the target component type of the photovoltaic power station based on the type parameters corresponding to the standard candidate bounding boxes and the type matching table; if the target component type of the photovoltaic power station is a PV panel, calculating the intersection-union ratio (IUGR) of the standard candidate bounding box and the upper half of the surface image of the target component; if the IUGR is 0, there is no PV component on the surface image of the target component, and the anti-fall strategy of the photovoltaic cleaning robot is triggered; if the IUGR is greater than 0, there is a PV component on the surface image of the target component, and the photovoltaic cleaning robot continues to move; if the target component recognition model of the photovoltaic power station outputs an unrecognizable signal, the anti-fall strategy of the photovoltaic cleaning robot is triggered based on the unrecognizable signal.

2. The identification method based on the target component identification model of a photovoltaic power station according to claim 1, characterized in that, The acquisition of surface images of target components in a photovoltaic power station includes: The camera on the photovoltaic cleaning robot is triggered when the robot moves relative to the photovoltaic panel. The camera takes pictures of the surface of the target components of the photovoltaic power plant as the photovoltaic cleaning robot moves relative to the photovoltaic panels.

3. The identification method based on the target component identification model of a photovoltaic power station according to claim 1, characterized in that, The step involves inputting the surface image of the target component of the photovoltaic power station into the target component recognition model of the photovoltaic power station. This target component recognition model is trained based on previous surface images and type parameters of target components from photovoltaic power stations, and includes: The target component preprocessing image of the photovoltaic power station is determined based on the surface image of the photovoltaic power station. The surface image of the target component of the photovoltaic power station is preprocessed to remove overexposed, underexposed and blurred images. The preprocessed image of the target component of the photovoltaic power station is input into the target component recognition model of the photovoltaic power station, and the target component recognition model of the photovoltaic power station identifies the component type. The target component recognition model of the photovoltaic power station is trained based on the surface images and type parameters of the target components of previous photovoltaic power stations.

4. The identification method based on the target component identification model of a photovoltaic power station according to claim 3, characterized in that, The step of inputting the surface image of the target component of the photovoltaic power station into the target component recognition model of the photovoltaic power station, wherein the target component recognition model of the photovoltaic power station is trained based on the surface images and type parameters of the target components of previous photovoltaic power stations, and further includes: To train the target component identification model for photovoltaic power plants, surface images of target components from multiple previous photovoltaic power plants were collected. An image dataset was created based on surface images of target components from multiple previous photovoltaic power plants; The image dataset is labeled to determine the type parameters of each surface image; The image dataset is divided into a training set, a validation set, and a test set, with a certain ratio between the training set, the validation set, and the test set. Collect known target component recognition models; The known target component recognition model learns its model parameters on the training set to complete the training of the known target component recognition model on the training set in the current round; After the known target component recognition model completes training on the training set in the current round, the known target component recognition model performs model validation and iterative training of hyperparameters on the validation set. The hyperparameters include learning rate, network width and depth, and various training weight parameters. After training and optimizing all hyperparameters of the known target component recognition model, a generalization ability test is performed on the test set to estimate the generalization error of the known target component recognition model. If the generalization error value is within the preset generalization error threshold, the identification model of the target component of the photovoltaic power station will be output.

5. The identification method based on the target component identification model of a photovoltaic power station according to claim 1, characterized in that, The target component type of the photovoltaic power station is defined according to the type parameter and the type matching table. The target component type of the photovoltaic power station includes photovoltaic components and non-photovoltaic components. Photovoltaic components include photovoltaic panels, briquettes, or mounting brackets, and also include: The formula for calculating the intersection-over-union ratio (IoU) of the standard candidate bounding box and the upper half of the surface image of the target module of the photovoltaic power plant is as follows: in, Indicates the height of the intersecting part; Indicates the width of the intersecting portion; Represents the area of ​​the intersection; This means taking the larger of the two. This means taking the minimum of the two. This represents the corresponding absolute value. For intersection, union, and comparison.

6. A recognition system based on a target component recognition model of a photovoltaic power station, characterized in that, The identification system based on the target component identification model of a photovoltaic power station is applied to the identification method based on the target component identification model of a photovoltaic power station as described in any one of claims 1-5, wherein the identification system based on the target component identification model of a photovoltaic power station includes: The acquisition module is used to acquire surface images of target components in a photovoltaic power plant. The input module is used to input the surface image of the target component of the photovoltaic power station into the target component recognition model of the photovoltaic power station. The target component recognition model of the photovoltaic power station is trained based on the surface images and type parameters of the target components of previous photovoltaic power stations. The output module is used to output the corresponding type parameters based on the target component recognition model of the photovoltaic power station. This includes: dividing the surface image of the target component of the photovoltaic power station into multiple candidate bounding boxes, where each candidate bounding box is labeled with its corresponding type parameter; performing bounding box confidence filtering on the multiple candidate bounding boxes to determine the candidate bounding boxes to be tested; calculating the distributed focus loss based on the candidate bounding boxes to determine the focus loss value of each candidate bounding box; if the focus loss value of a candidate bounding box to be tested is less than a preset focus loss threshold, then the candidate bounding box to be tested is determined as a standard candidate bounding box, and the type parameter corresponding to the standard candidate bounding box is determined. The coordinates of the standard bounding boxes are obtained through distributed focus loss. The confidence of the bounding boxes is dequantized, and the confidence and classification results are recorded. The bounding boxes obtained under different time lengths are summed, and these bounding boxes are quickly sorted according to their confidence. The non-maximum suppression algorithm is used to select the bounding boxes with high confidence in the overlapping area that have a confidence greater than the threshold as the result output. The structure of the distributed focus loss function is moved outside the target component recognition model of the photovoltaic power station. The distributed focus loss is calculated on the CPU. The target component type module is used to define the target component type of a photovoltaic power station based on type parameters and a type matching table. The target component type of the photovoltaic power station includes photovoltaic modules and non-photovoltaic modules. Photovoltaic modules include photovoltaic panels, clamps, or mounting brackets. The module includes: collecting type parameters corresponding to standard candidate bounding boxes; associating the type parameters corresponding to the standard candidate bounding boxes with the type matching table; determining the target component type of the photovoltaic power station based on the type parameters corresponding to the standard candidate bounding boxes and the type matching table; if the target component type of the photovoltaic power station is a photovoltaic panel, calculating the intersection-union ratio (IUGR) of the standard candidate bounding box and the upper half of the surface image of the target component of the photovoltaic power station; if the IUGR is 0, there is no photovoltaic component on the surface image of the target component of the photovoltaic power station, and the anti-fall strategy of the photovoltaic cleaning robot is triggered; if the IUGR is greater than 0, there is a photovoltaic component on the surface image of the target component of the photovoltaic power station, and the photovoltaic cleaning robot continues to move; if the target component recognition model of the photovoltaic power station outputs an unrecognizable signal, the anti-fall strategy of the photovoltaic cleaning robot is triggered based on the unrecognizable signal.

Images