Defect detection method and device for photovoltaic array polling by unmanned aerial vehicle, electronic equipment and storage medium
Through the cascading image segmentation and defect detection model, combined with the coordinate attention mechanism, the problem of precise positioning and efficiency improvement in the defect detection of drone photovoltaic arrays is solved, and the support of safety and data-driven decisions is achieved, and the photovoltaics is improved. Array inspection efficiency and safety.
Patent Information
- Application Number
- CN202410931689.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-12
- Publication Date
- 2024-11-12
AI Technical Summary
The existing drone photovoltaic array defect detection technology has shortcomings in precise positioning and efficiency improvement, and safety and data-driven decision-making support need to be improved.
The image segmentation model and defect detection model are adopted in cascading settings. By acquiring infrared images and visible light images, the photovoltaic components are generated and intermediate processing is performed. The defect detection model is optimized in combination with the coordinate attention mechanism to realize the post-processing of defect detection results and Alarm.
It improves the efficiency of photovoltaic array inspection, achieves accurate positioning of faults, improves inspection safety, and supports digital management through deep learning technology to achieve customized alarms and precise positioning.
Smart Images

Figure CN118941507A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of specific computer models for intelligent inspection, and more specifically, relates to a defect detection method, device, electronic equipment and storage medium for unmanned aerial vehicle inspection of photovoltaic arrays. Background Art
[0002] Photovoltaic array defect detection by drone is an innovative practice that deeply combines the current smart grid and new energy fields. It uses advanced drones equipped with high-precision imaging equipment and deep learning algorithms to achieve efficient and accurate inspections of large-scale photovoltaic power stations. At present, this technology has shown significant application prospects and important practical value worldwide.
[0003] In terms of the current state of technology, the drone photovoltaic inspection system integrates remote sensing technology, image processing technology and artificial intelligence technology. The drone can fly according to a preset or autonomously planned route, and obtain high-definition image data of the photovoltaic panel through multimodal sensors such as visible light cameras and infrared thermal imagers. These data are then wirelessly transmitted to the ground station for analysis and processing. Deep learning algorithms play a key role in this link. By learning from a large number of labeled samples, the model can automatically identify various photovoltaic panel defects including dust accumulation, pollutant occlusion, vegetation occlusion, etc., and can even detect in real time the problem of reduced power generation efficiency caused by shadow occlusion, hot spot effect, etc.
[0004] The significance of drone photovoltaic panel defect detection is mainly reflected in the following aspects:
[0005] Improve inspection efficiency: Compared with traditional manual inspection methods, drones can quickly cover large areas and are not restricted by terrain, which greatly shortens the inspection cycle and reduces labor costs.
[0006] Accurately locate faults: Deep learning methods can accurately identify subtle defects, helping to promptly discover and locate problem points, thereby reducing potential power generation losses and ensuring the overall operating efficiency of the power station.
[0007] Enhanced safety: Drones do not require personnel to climb to high altitudes for operation, effectively avoiding personal safety risks for operation and maintenance personnel.
[0008] Data-driven decision-making: Big data collected by drone inspections can support long-term performance evaluation and maintenance strategy optimization, providing a solid data foundation for intelligent operation and maintenance.
[0009] It can be seen that the use of drones equipped with deep learning technology to detect photovoltaic panel defects is an important breakthrough in modern clean energy management technology, and has far-reaching significance for improving the operation and maintenance level of solar power stations, ensuring a stable supply of energy, and promoting green and sustainable development.
[0010] Chinese patent document CN116309293A discloses a photovoltaic module defect detection method, device, electronic device and storage medium, the method comprising: obtaining a visible light image and an infrared image of a target photovoltaic module; based on camera calibration parameters, converting the visible light image and the infrared image to the same target plane; fusing the visible light image and the infrared image in the target plane to obtain a fused image; performing defect detection based on an image segmentation model and the fused image to obtain a defect detection result of the target photovoltaic module. By converting the visible light image and the infrared image to the same target plane, the visible light image and the infrared image can be fused in the target plane to obtain a fused image, and performing defect detection based on the fused image can improve the accuracy of photovoltaic module defect detection.
[0011] Chinese patent document CN116385421A discloses a photovoltaic panel detection method, a drone and a computer-readable storage medium, wherein the photovoltaic panel detection method comprises: obtaining a time-series multimodal image collected by a drone, and obtaining a visible light image and an infrared image corresponding to the photovoltaic panel to be detected based on the time-series multimodal image; extracting common features of the visible light image and the infrared image, and generating a fused image corresponding to the visible light image and the infrared image based on the common features; semantically segmenting the fused image based on the photovoltaic panel features, and inputting the fused image into an improved defect detection model, and increasing the weight value of the detection target in the fused image based on the attention mechanism to determine the segmented area and defect type where the abnormality is located in the fused image; determining the abnormal position according to the segmented area, and outputting the photovoltaic panel detection result based on the defect type and the abnormal position. Through drone inspection, automatic identification and positioning of photovoltaic panel defects and abnormalities are realized, and the efficiency of abnormality detection of large-volume photovoltaic panels is improved.
[0012] In view of this, the present invention designs a defect detection method for unmanned aerial vehicle inspection of photovoltaic arrays, which is used to improve the efficiency of inspection of photovoltaic arrays, accurately locate faults, and improve inspection safety. Summary of the invention
[0013] The present invention aims to overcome at least one defect of the above-mentioned prior art and provide a defect detection method for unmanned aerial vehicle inspection of photovoltaic arrays.
[0014] The present invention also discloses a device equipped with a defect detection method for unmanned aerial vehicle inspection of photovoltaic arrays.
[0015] The detailed technical scheme of the present invention is as follows:
[0016] A method for detecting defects in a photovoltaic array by using an unmanned aerial vehicle for inspection, the method comprising:
[0017] S1, obtaining an infrared image and a visible light image of each photovoltaic array, and pairing the infrared image and the visible light image of each photovoltaic array to form a plurality of groups of images to be detected;
[0018] S2, inputting the image to be detected into a photovoltaic array defect detection model, the photovoltaic array defect detection model includes a segmentation model and a defect detection model arranged in cascade, the image to be detected is segmented by the segmentation model to generate a number of photovoltaic modules, the photovoltaic modules are input into the defect detection model after intermediate processing, and the corresponding defect detection results are output, the defect detection results include defect categories and category confidence;
[0019] The segmentation model includes an infrared image segmentation model for segmenting infrared images and a visible light image segmentation model for segmenting visible light images, the defect detection model includes an infrared image defect detection model for identifying infrared images and a visible light image defect detection model for identifying visible light images, and the infrared image segmentation model and the infrared image defect detection model are cascaded, and the visible light image segmentation model and the visible light image defect detection model are cascaded;
[0020] S3. Post-process the defect detection result of the image to be detected to determine whether it is a defective image. If so, further determine whether it is an infrared image defect or a visible light image defect.
[0021] Preferably, according to the present invention, in step S2, the segmentation model adopts the Yolov8l-seg-640 model, and the defect detection model adopts the Yolov8l-dec-640 model optimized by the coordinate attention mechanism, that is, a three-layer attention module CA is inserted into the Yolov8l-dec-640 model, and the three-layer attention module CA serves as the next layer structure of the second C2F layer, the next layer structure of the third C2F layer, and the next layer structure of the fourth C2F layer in the Yolov8l-dec-640 model structure, respectively.
[0022] Preferably, according to the present invention, step S2 further includes: constructing a segmentation data set and a defect detection data set based on the training data set to train the segmentation model and the defect detection model respectively, wherein the training data set includes training samples formed by pairing an infrared image and a visible light image of each photovoltaic array;
[0023] The segmentation dataset and the defect detection dataset are constructed based on the training dataset, specifically including:
[0024] Selecting some training samples from the training data set, and manually labeling all complete photovoltaic components in the entire image of each training sample to form segmentation data, thereby constructing a segmentation data set for training a segmentation model;
[0025] Cutting the training samples that have been labeled with all the complete photovoltaic modules, and cutting the remaining unlabeled training samples in the training data set using the trained segmentation model to form a number of photovoltaic module samples;
[0026] The defect locations in all photovoltaic module samples are marked by manual labeling to form defect detection data, thereby constructing a defect detection dataset for training defect detection models.
[0027] Preferably, according to the present invention, in step S2, the photovoltaic module is subjected to intermediate processing, specifically comprising:
[0028] Performing morphological operations of corrosion and expansion on the photovoltaic modules segmented in the image to be detected, so that the segmented edges are connected to form a number of connected domains; wherein, for the photovoltaic modules in the visible light image, a 23*23 structural element is used for expansion; and for the photovoltaic modules in the infrared image, a 5*5 structural element is used for expansion;
[0029] Determine the number of connected domains formed in the image to be detected after expansion:
[0030] If the number of connected domains is one, the connected domain is determined as the photovoltaic string to be tested;
[0031] If the number of connected domains is greater than one, the distance between the centroid of each connected domain and the center point of the image to be detected is calculated, and the connected domain where the centroid of the minimum distance is located is determined as the photovoltaic string to be tested;
[0032] The step of calculating the distance between the centroid of each connected domain and the center point of the image to be detected to determine the photovoltaic string to be detected specifically includes:
[0033] Use opencv software to obtain the height H and width W of the image to be detected, and use (W / 2, H / 2) as the center point coordinates of the image to be detected;
[0034] The number of all connected domains in the image to be detected and the coordinate information of each connected domain are obtained by the cv2.connectedComponentsWithStats() method, and the height and width of each connected domain are determined according to the coordinate information of each connected domain, so as to calculate the centroid coordinates of each connected domain;
[0035] The distance between the coordinates of the centroid of each connected domain and the coordinates of the center point of the image to be detected is calculated by the Pythagorean theorem, and the connected domain where the centroid of the minimum distance is located is determined as the photovoltaic string to be detected.
[0036] Preferably, according to the present invention, in step S2, the photovoltaic module is subjected to intermediate processing, which specifically further includes:
[0037] The highest point and the lowest point of the photovoltaic string to be tested are obtained according to the bounding box of the connected domain, and half of the distance between the highest point and the lowest point is used as the height of each photovoltaic module in the photovoltaic string to be tested to determine the position of the upper and lower dividing lines of the photovoltaic string to be tested;
[0038] The row number to which each photovoltaic module belongs is determined based on the relationship between the y-value of the ordinate of each photovoltaic module and the ordinate value of the dividing line, if:
[0039] If the y-value of the ordinate of the photovoltaic module is less than the ordinate value of the dividing line, it belongs to the first row;
[0040] If the y-value of the ordinate of the photovoltaic module is greater than the ordinate value of the dividing line, it belongs to the second row;
[0041] The column number to which each photovoltaic module belongs is determined by the size of the x-coordinate of each photovoltaic module, and a set of serial numbers is formed as the serial number of the corresponding photovoltaic module;
[0042] The first photovoltaic module in the photovoltaic string to be tested has a serial number (1,1).
[0043] Preferably, according to the present invention, step S3 specifically includes:
[0044] S31, judging whether the defect detection result is empty, if it is empty, the image to be detected is determined to be a defect-free image; if it is not empty, entering S32;
[0045] S32, judging whether there is only one defect detection result and it is a normal result, if so, the image to be detected is determined to be a defect-free image; otherwise, entering S33;
[0046] S33, sorting the defect detection results according to the size of the category confidence. If there are multiple results, among which the confidence of the normal result is the largest and greater than 0.8, the image to be detected is considered to be a defect-free image; otherwise, entering S34;
[0047] S34. According to the defect categories contained in the infrared image and the visible light image, determine whether the defect detection result is an infrared image defect or a visible light image defect, wherein the defect categories contained in the infrared image include diode short circuit, shadow occlusion, vegetation occlusion and hot spot; the defect categories contained in the visible light image include pollutant occlusion, shadow occlusion, vegetation occlusion and severe dust accumulation.
[0048] Preferably, according to the present invention, the method further comprises: comparing the category confidence contained in the defect detection result with the corresponding preset threshold value, and triggering an alarm if the category confidence in the defect detection result is greater than the corresponding preset threshold value;
[0049] Among them, for the defect categories contained in the infrared image, the preset thresholds corresponding to the category confidence are: diode short circuit 0.7; shadow occlusion 0.6; vegetation occlusion 0.5; hot spot 0.6; for the defect categories contained in the visible light image, the preset thresholds corresponding to the category confidence are: pollutant occlusion 0.5; shadow occlusion 0.55; vegetation occlusion 0.5; severe dust accumulation 0.4.
[0050] In another aspect of the present invention, a device for implementing a defect detection method for a photovoltaic array inspected by a drone is provided, the device comprising:
[0051] A data acquisition module is used to acquire the infrared image and visible light image of each photovoltaic array, and pair the infrared image and visible light image of each photovoltaic array to form a plurality of groups of images to be detected;
[0052] An image segmentation and defect detection module is used to input the image to be detected into a photovoltaic array defect detection model, the photovoltaic array defect detection model includes a segmentation model and a defect detection model arranged in cascade, the image to be detected is segmented by the segmentation model to generate a number of photovoltaic modules, the photovoltaic modules are input into the defect detection model after intermediate processing, and the corresponding defect detection results are output, the defect detection results include defect categories and category confidence;
[0053] The segmentation model includes an infrared image segmentation model for segmenting infrared images and a visible light image segmentation model for segmenting visible light images, the defect detection model includes an infrared image defect detection model for identifying infrared images and a visible light image defect detection model for identifying visible light images, and the infrared image segmentation model and the infrared image defect detection model are cascaded, and the visible light image segmentation model and the visible light image defect detection model are cascaded;
[0054] a detection result post-processing module, used to post-process the defect detection result of the image to be detected to determine whether it is a defective image, and if so, further determine whether it is an infrared image defect or a visible light image defect; and
[0055] The alarm module is used to compare the category confidence contained in the defect detection result with the corresponding preset threshold, and trigger an alarm if the category confidence in the defect detection result is greater than the corresponding preset threshold.
[0056] In another aspect of the present invention, there is also provided an electronic device, comprising:
[0057] at least one processor; and
[0058] A memory storing instructions, which, when executed by the at least one processor, enables the at least one processor to execute the defect detection method for unmanned aerial vehicle inspection of photovoltaic arrays as described above.
[0059] In another aspect of the present invention, a machine-readable storage medium is provided, which stores executable instructions, and when the instructions are executed, the machine executes the defect detection method for unmanned aerial vehicle inspection of photovoltaic arrays as described above.
[0060] Compared with the prior art, the present invention has the following beneficial effects:
[0061] (1) The present invention provides a defect detection method for unmanned aerial vehicle inspection of photovoltaic arrays. By establishing a cascaded image segmentation model and an image defect detection model, the image to be inspected is segmented into several photovoltaic components for defect detection. The defect detection results are then post-processed and analyzed. Finally, the defect information is displayed from two dimensions, including aggregating and merging the detection results of all photovoltaic components, and finally returning the defects contained in the entire image, as well as displaying the location information and defect information of the defective photovoltaic components, so as to display the specific location of the defect. The method improves the efficiency of photovoltaic array inspection, realizes accurate fault positioning, and improves inspection safety.
[0062] (2) The present invention combines drone inspection technology with deep learning technology to achieve digital management of photovoltaic power stations. Through the cascade application of deep models and customized post-processing methods, defects can be accurately located on each photovoltaic panel, and customized alarms can be implemented according to actual needs. Defect results can be displayed and alarms can be issued on the platform side. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] Figure 1 It is a flow chart of the defect detection method of the unmanned aerial vehicle inspection photovoltaic array described in the present invention.
[0064] Figure 2 It is a flowchart of the recognition logic of the photovoltaic array defect detection model and the post-processing of the detection results in Example 1 of the present invention.
[0065] Figure 3 It is a schematic diagram of the network structure of the defect detection model in Example 1 of the present invention.
[0066] Figure 4 is a diagram showing the segmentation results in Example 1 of the present invention, wherein: Figure 4 (a) is the segmentation result display of the infrared image. Figure 4 (b) is a diagram showing the segmentation results of the visible light image.
[0067] Figure 5 is a diagram showing the expansion results in Example 1 of the present invention, wherein: Figure 5 (a) is the expansion result display of the infrared image. Figure 5 (b) is a diagram showing the expansion results of the visible light image.
[0068] Figure 6 is a display diagram of the positioning of the photovoltaic string to be tested in Example 1 of the present invention, wherein: Figure 6 (a) is the infrared image showing the location of the photovoltaic string to be tested. Figure 6 (b) is a visible light image showing the positioning of the photovoltaic string to be tested.
[0069] Figure 7 is a diagram showing the numbering of photovoltaic modules in Example 1 of the present invention, wherein: Figure 7 (a) is a display of the photovoltaic module numbering in the infrared image. Figure 7 (b) is a visible light image showing the PV module numbers.
[0070] Figure 8 It is a result display diagram of the defect detection result after post-processing in the present invention, wherein: Figure 8 (a) is an infrared image showing defects of a single photovoltaic module; Figure 8 (b) is an infrared image showing defects in photovoltaic strings; Figure 8 (c) is a visible light image showing defects of a single photovoltaic module; Figure 8 (d) is a visible light image showing photovoltaic string defects. DETAILED DESCRIPTION
[0071] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.
[0072] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.
[0073] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0074] In the absence of conflict, the embodiments in the present disclosure and the features in the embodiments may be combined with each other.
[0075] In view of the shortcomings in the prior art, the present invention provides a defect detection method for photovoltaic arrays during drone inspection, which involves the fields of deep learning, remote sensing technology, and intelligent patrol. It includes acquiring photovoltaic array image data taken by a drone, including infrared images and visible light images of each photovoltaic array, pairing them to form a plurality of groups of images to be detected, and then transmitting the images to be detected to an analysis host. The plurality of groups of images to be detected are sequentially input into a pre-trained photovoltaic array defect detection model for detection and analysis, and the detection results are post-processed. If it is determined that the image has abnormal defects, an alarm and a display are issued.
[0076] The present invention combines drone inspection technology with deep learning technology to achieve digital management of photovoltaic power stations. Through the cascade application of deep models and customized post-processing methods, defects can be accurately located on each photovoltaic panel, and customized alarms can be implemented according to actual needs. Defect results can be displayed and alarms can be issued on the platform side.
[0077] The defect detection method and device for unmanned aerial vehicle inspection of photovoltaic arrays of the present invention are further described below in conjunction with specific embodiments.
[0078] Embodiment 1,
[0079] Ginseng Figure 1 , this embodiment provides a defect detection method for a photovoltaic array by using a drone for inspection, the method comprising:
[0080] S1. Acquire an infrared image and a visible light image of each photovoltaic array, and pair the infrared image and the visible light image of each photovoltaic array to form a plurality of groups of images to be detected.
[0081] Specifically, the image data of the photovoltaic array can be collected by drone inspection, including infrared images and visible light images of each photovoltaic array. In this embodiment, the inspection drone model is DJI Mavic 3T, and its parameter configuration includes: wide angle: equivalent focal length 24 mm, 48 million pixels, telephoto: equivalent focal length 162 mm, 12 million pixels, 56 times hybrid zoom, thermal imaging: DFOV: 61°, equivalent focal length 40 mm, resolution 640*512. During shooting, infrared images and visible light images are shot in pairs.
[0082] Then, the infrared image and visible light image of each photovoltaic array are paired to form multiple groups of images to be detected. Specifically, the file names can be matched in pairs, and the infrared image is named: timestamp_T.JPG; the visible light image is named: timestamp_V.JPG. The infrared image and the visible light image are paired one by one, and finally the infrared image and the visible light image of a photovoltaic array are paired to form a group of images to be detected.
[0083] S2. Input the image to be detected into a photovoltaic array defect detection model, wherein the photovoltaic array defect detection model includes a segmentation model and a defect detection model arranged in cascade. The image to be detected is segmented by the segmentation model to generate a number of photovoltaic components. The photovoltaic components are input into the defect detection model after intermediate processing, and the corresponding defect detection results are output, wherein the defect detection results include defect categories and category confidence.
[0084] In this embodiment, the segmentation model and the defect detection model can both adopt the open source algorithm Yolov8 series model, wherein the segmentation model can adopt the Yolov8l-seg-640 model, and the defect detection model can adopt the Yolov8l-dec-640 model optimized by the coordinate attention mechanism.
[0085] Due to the regularity of the target to be segmented, the Yolov8l-seg-640 model can achieve the expected segmentation effect, so no model optimization is performed on it. Since the defect targets contain pollutants (mostly bird excrement), small targets such as hot spots, and occlusion, the attention mechanism is needed to optimize the recognition effect of the defect detection model.
[0086] By optimizing the coordinate attention mechanism, we can enhance the sensitivity to position information, improve feature expression capabilities, and reduce computing resource consumption. Figure 3 , a three-layer CA (CoordinateAttention) structure is added to the backbone network of Yolov8, which is placed behind the 2nd, 3rd, and 4th C2F layers of the original structure. As an important means to enhance the expressiveness of the model, the working principle of the coordinate attention mechanism can be summarized as follows: First, it receives any intermediate tensor as input, which contains image features and whose dimensions are defined by the number i, height H, width W, and number of channels C.
[0087] To obtain the attention information of the image in the width W and height H directions, the mechanism first performs the following operations:
[0088] ① Average pooling: Perform global average pooling on the input feature map in both width W and height H directions, so that feature maps in both directions can be obtained.
[0089] ② Feature map concatenation: Concatenate the feature maps in the two directions obtained above to form a new feature map. This new feature map contains information in both the width W and height H directions.
[0090] ③ Dimensionality reduction: Input the concatenated new feature map into a 1×1 convolution module for dimensionality reduction to reduce the dimension of the feature map. After this process, the dimension of the feature map will become 1×(W+H)×C / r, where C / r represents the dimensionality reduction multiple of the feature map in the direction of the number of channels C.
[0091] ④ Weight generation: The reduced feature map will then pass through the batch normalization layer and the Sigmoid activation function, so that the attention weights in the width W and height H directions can be obtained. At the same time, in order to restore the original dimensions of the feature map in these two directions, the feature map will pass through two parallel 1×1 convolution modules and apply the Sigmoid activation function respectively, so as to obtain the attention weights in the width W and height H directions.
[0092] ⑤ Attention weighting: Finally, the obtained attention weights will be multiplied and weighted (Re-weight) with the original image features of the model input to realize the attention mechanism in the two directions of width W and height H. This weighted operation can highlight the important parts of the image and suppress the unimportant parts, thereby enhancing the model's ability to express and understand image features.
[0093] After the above model is built, it is necessary to pre-train the constructed model. The data used for training is also the photovoltaic array image data collected by drone inspection, including infrared images and visible light images of each photovoltaic array. The infrared images and visible light images are then paired one by one to form training samples. The paired training samples are then segmented and labeled for defect detection to construct a training data set, which is used to train the segmentation model and defect detection model respectively.
[0094] Since it is difficult to label the training data of the segmentation model and the amount of data used for training is less than that of the defect detection model, some training samples are first selected from the constructed training data set during labeling, and they are segmented and labeled by manual labeling as the segmentation data for training the segmentation model.
[0095] When annotating the segmentation data for training the segmentation model, it is necessary to annotate all the complete photovoltaic modules in the entire image to form a segmentation data set. After there is enough segmentation data, the segmentation model can be trained. The model training process is a conventional technology and will not be described in detail here.
[0096] The trained segmentation model is used to crop and segment the photovoltaic array image (including infrared images and visible light images) into several photovoltaic components.
[0097] The defect detection model and the segmentation model are cascaded, and the output of the segmentation model will be used as the input of the defect detection model. Therefore, the training data of the defect detection model is the cropped PV modules. When the effect of the segmentation model reaches the expected effect, the trained segmentation model can be used to process the training data required by the defect detection model.
[0098] That is, first, part of the original segmented and labeled training samples are cropped according to the segmentation labels to form several small images, namely, photovoltaic module samples. The amount of data at this time is far from sufficient to support the training of the defect detection model. It is necessary to use the trained segmentation model to segment the remaining unlabeled training samples in the training data set, and segment and crop a sufficient amount of training data required by the defect detection model. Then, the segmented training data are manually labeled for defect detection as the defect detection data for training the defect detection model.
[0099] When annotating the defect detection data for training the defect detection model, it is necessary to annotate the specific locations of the defects on all the segmented photovoltaic module samples. Defects include hot spots, diode short circuits, shadow occlusion, pollutant occlusion, vegetation occlusion, severe dust accumulation, module breakage, module detachment, and module non-power generation. Finally, a defect detection data set is formed. After there is enough defect detection data, the defect detection model is trained. The model training process is a conventional technology and will not be described in detail here.
[0100] The trained defect detection model is used to detect whether there are defects in photovoltaic modules (including photovoltaic modules in infrared images and photovoltaic modules in visible light images) and output corresponding defect detection results.
[0101] Furthermore, the segmentation model in this embodiment includes an infrared image segmentation model for segmenting infrared images and a visible light image segmentation model for segmenting visible light images, and the defect detection model includes an infrared image defect detection model for identifying infrared images and a visible light image defect detection model for identifying visible light images, and the infrared image segmentation model and the infrared image defect detection model are cascaded, and the visible light image segmentation model and the visible light image defect detection model are cascaded. The two sets of models have the same recognition logic, but different result processing. When performing defect detection, a set of images to be detected respectively call the corresponding cascade algorithm.
[0102] Ginseng Figure 2 , which shows the recognition logic of the photovoltaic array defect detection model in this embodiment.
[0103] Specifically, a set of images to be detected (paired infrared images and visible light images) is input into the trained photovoltaic array defect detection model. The images to be detected are first segmented and cropped by the trained segmentation model to output a number of photovoltaic components. Among them, the infrared image is segmented and cropped by its corresponding infrared image segmentation model to output a number of infrared images of photovoltaic components; the visible light image is segmented and cropped by its corresponding visible light image segmentation model to output a number of visible light images of photovoltaic components. The image segmentation result is referenced. Figure 4 As shown, Figure 4 (a) is the segmentation result display of the infrared image. Figure 4 (b) is a diagram showing the segmentation results of the visible light image.
[0104] Since the focal lengths of infrared images and visible light images are different, visible light images are wide-angle shots, while infrared images are telephoto shots. Therefore, most of a set of images can only cover a complete set of photovoltaic arrays (2*14 photovoltaic modules), and the segmentation results need to be processed, that is, all generated photovoltaic modules are intermediately processed. The segmentation results are input into the segmentation data post-processing module, and the corresponding complete photovoltaic arrays in the infrared image and visible light image are located through the positioning and sorting algorithm. This array is the array to be tested (i.e., the photovoltaic string to be tested), and each photovoltaic module is given a number for subsequent defect merging of infrared images and visible light images and position positioning during display. Specifically:
[0105] The generated photovoltaic modules are subjected to morphological operations of corrosion and expansion to connect the segmented edges and form several connected domains. A group of photovoltaic modules forms a connected domain. For the photovoltaic modules in the visible light image, a 23*23 structural element is used for expansion; for the photovoltaic modules in the infrared image, a 5*5 structural element is used for expansion.
[0106] The dilation operation is mainly used to expand or dilate an image. The basic idea is to use a structural element to scan the image, and replace the image value corresponding to the center point of the structural element each time the scan is performed. The formula is:
[0107] G(x,y)=F(x+s,y+d),(s,d)∈B, where G represents the dilation result, i.e. the new image generated by the dilation; F represents the object to be dilated, i.e. the image before dilation; B represents the structural element; (x,y) is any pixel on the image; (s,d) is any point in the structural element. For each pixel (x,y) in the image F, cover it with the structural element B, and then take the maximum value of all pixels within the coverage range of the structural element B as the value of the corresponding position in the new image G.
[0108] Through the above operation, the gaps between photovoltaic modules can be filled, so that a group of photovoltaic modules form a connected domain. Figure 5 As shown, Figure 5 (a) is the expansion result display of the infrared image. Figure 5 (b) is a diagram showing the expansion results of the visible light image.
[0109] After obtaining the connected domain, determine how many connected domains there are in the expanded image to be detected. If there is only one connected domain, determine that the connected domain is the photovoltaic string to be tested; if there are two or more connected domains, calculate the distance between the centroid of each connected domain and the center point of the image to be detected, and determine the connected domain where the minimum distance centroid is located as the photovoltaic string to be tested, that is, determine the photovoltaic string at the center of the acquired image to be detected as the photovoltaic string to be tested. This is because there may be incomplete connected domains in the image to be detected. For example, the connected domain at the edge of the image may only show part of it. Therefore, the connected domain in the middle of the image can be obtained through the centroid to determine whether it is a complete connected domain in the entire image, so as to determine the photovoltaic string to be tested.
[0110] The specific implementation of this process is as follows: 1) Use opencv to read the image to be detected, obtain the height H and width W of the image, and use (W / 2, H / 2) as the centroid coordinates of the image; 2) Use the cv2.connectedComponentsWithStats() method to obtain the number, coordinates and other related information of all connected domains in the image to be detected; 3) Determine the height and width of each connected domain based on the obtained coordinate information of each connected domain, and calculate the centroid coordinates of each connected domain; 4) Use the Pythagorean theorem to calculate the distance between the centroid coordinates of each connected domain and the centroid coordinates of the image to be detected, and take the connected domain with the minimum distance centroid as the center position of the photovoltaic string, that is, the photovoltaic string to be tested. The positioning result of the photovoltaic string to be tested is referenced. Figure 6 As shown, Figure 6 (a) is the infrared image showing the location of the photovoltaic string to be tested. Figure 6 (b) is a visible light image showing the positioning of the photovoltaic string to be tested.
[0111] After obtaining the photovoltaic string to be tested, it is necessary to sort and number all the photovoltaic modules in the photovoltaic string to be tested. The sorting is done by row. Each complete photovoltaic array is generally composed of 2*14 photovoltaic modules (may be different in different sites). The photovoltaic modules are sorted by row. For example, the first photovoltaic module in the upper left corner is numbered (1, 1), which means that its position is the first row and the first column. The highest and lowest points of the area to be tested (i.e., the photovoltaic string to be tested obtained above) are obtained through the connected domain boundary box. One-half of the distance between the two points is used as the height of the photovoltaic module to determine the position of the upper and lower dividing lines of the area to be tested. The row number to which the photovoltaic module belongs is determined by the size relationship between the y value of the ordinate of the photovoltaic module and the ordinate of the dividing line. If the y value of the ordinate of the photovoltaic module is less than the ordinate of the dividing line, it belongs to the first row; if the y value of the ordinate of the photovoltaic module is greater than the ordinate of the dividing line, it belongs to the second row; then the number of its columns is determined by the size of its abscissa x, and finally the first photovoltaic module in the first row (i.e., the first photovoltaic module in the upper left corner) is numbered (1, 1), and so on. The sorting and numbering results of all photovoltaic modules are referenced. Figure 7 As shown, Figure 7 (a) is a display of the photovoltaic module numbering in the infrared image. Figure 7 (b) is a visible light image showing the PV module numbers.
[0112] All photovoltaic modules after intermediate processing are assigned corresponding serial numbers. Finally, these serial numbered images are input into the trained defect detection model for defect identification. The model outputs the defect detection results corresponding to each photovoltaic module. If the result is not empty, the result contains the defect category and category confidence corresponding to each photovoltaic module.
[0113] S3. Post-process the defect detection result of the image to be detected to determine whether it is a defective image. If so, further determine whether it is an infrared image defect or a visible light image defect.
[0114] In this embodiment, first, it is set that for infrared image defects, the thresholds of defect categories and category confidences are set as follows: diode short circuit: 0.7; shadow occlusion: 0.6; vegetation occlusion: 0.5; hot spot: 0.6; for visible light image defects, the thresholds of defect categories and category confidences are set as follows: pollutant occlusion: 0.5; shadow occlusion: 0.55; vegetation occlusion: 0.5; severe dust accumulation: 0.4; in addition, pollutant occlusion needs to be judged separately.
[0115] When judging whether the image to be detected is a defective image according to the defect detection result of the image to be detected, firstly judge whether the defect detection result is empty. If it is empty, the image to be detected is determined to be a defect-free image; if it is not empty, proceed to the next step of judgment.
[0116] It is determined whether there is only one defect detection result and it is a normal result. If so, the image to be detected is determined to be a defect-free image; otherwise, the next step of judgment is carried out.
[0117] The defect detection results are sorted according to the size of the category confidence. If there are multiple results, but the confidence of the normal result is the largest and greater than 0.8, the image to be detected is considered to be a defect-free image; otherwise, the next step of judgment is carried out.
[0118] For example, the defect detection results of a photovoltaic module are (severe dust accumulation, confidence 0.5), (contaminant occlusion, confidence 0.2), and (normal, confidence 0.89). Since the confidence of severe dust accumulation and contaminant occlusion is low and the confidence of the normal category is the highest and greater than 0.8, it can be considered that severe dust accumulation and contaminant occlusion are false detections, and the image is a defect-free image, that is, a normal image; when the defect detection results are (severe dust accumulation, confidence 0.8), (contaminant occlusion, confidence 0.2) (normal, confidence 0.4), the normal confidence is low at this time, it can be considered that the normal result may be a false detection, and the judgment can continue according to the conventional defect identification process.
[0119] The defect detection result includes defect category information. By combining the defect categories included in the infrared image and the visible light image, it can be determined whether the defective image is an infrared image defect or a visible light image defect.
[0120] After the post-processing of the results is completed, the output results are transmitted to the platform side for display. The display is displayed in two dimensions: one is the defects of the entire photovoltaic array, that is, the inspection results of all photovoltaic modules are summarized and merged, and finally the defects contained in the entire photovoltaic array image are returned; the other is the defects of all photovoltaic modules on the photovoltaic array, including their location information and defect information, which are used to display the specific location of the positioning defects. By linking with the front end, it can be displayed in layers according to needs. The results are displayed as follows Figure 8 As shown, it includes component defects and photovoltaic array defects, among which, Figure 8 (a) is an infrared image showing defects of a single photovoltaic module; Figure 8 (b) is an infrared image showing defects in photovoltaic strings; Figure 8 (c) is a visible light image showing defects of a single photovoltaic module; Figure 8 (d) is a visible light image showing photovoltaic string defects.
[0121] Furthermore, this embodiment customizes the alarm strategy according to actual needs, that is, the known defect detection results contain the category confidence corresponding to the defect category, and the category confidence is compared with its corresponding preset threshold. If the category confidence is greater than its corresponding preset threshold, the alarm condition is met and an alarm will be triggered.
[0122] Combining the defect categories of infrared images and visible light images, for defects such as diode short circuit, shadow occlusion, vegetation occlusion, hot spots and severe dust accumulation, an alarm will be issued as long as they exist in the detection results and the confidence level is greater than the set threshold. For the defect of pollutant occlusion, in addition to the judgment requirements for its confidence level, its area must also be judged. When the detected pollutant area is larger than the set threshold, an alarm will be issued, otherwise it will only be displayed.
[0123] The defect detection method for unmanned aerial vehicle inspection of photovoltaic arrays of the present invention realizes photovoltaic module defect detection through cascading segmentation and detection models. Compared with the detection of the entire photovoltaic array, it can display the defect position with finer granularity and is more intuitive for defect positioning. It also greatly reduces the difficulty of data acquisition for defect detection of photovoltaic modules. A large amount of defect training data can be cropped from one picture, thereby improving the recognition effect of defect detection.
[0124] Embodiment 2,
[0125] This embodiment provides a device for implementing a defect detection method for a photovoltaic array by drone inspection, the device comprising:
[0126] A data acquisition module is used to acquire the infrared image and visible light image of each photovoltaic array, and pair the infrared image and visible light image of each photovoltaic array to form a plurality of groups of images to be detected;
[0127] An image segmentation and defect detection module is used to input the image to be detected into a photovoltaic array defect detection model, the photovoltaic array defect detection model includes a segmentation model and a defect detection model arranged in cascade, the image to be detected is segmented by the segmentation model to generate a number of photovoltaic modules, the photovoltaic modules are input into the defect detection model after intermediate processing, and the corresponding defect detection results are output;
[0128] The segmentation model includes an infrared image segmentation model for segmenting infrared images and a visible light image segmentation model for segmenting visible light images, the defect detection model includes an infrared image defect detection model for identifying infrared images and a visible light image defect detection model for identifying visible light images, and the infrared image segmentation model and the infrared image defect detection model are cascaded, and the visible light image segmentation model and the visible light image defect detection model are cascaded;
[0129] a detection result post-processing module, used to post-process the defect detection result of the image to be detected to determine whether it is a defective image, and if so, further determine whether it is an infrared image defect or a visible light image defect, and obtain the defect category and category confidence corresponding to the infrared image defect or the visible light image defect; and
[0130] An alarm module is used to compare the category confidence of the defect category corresponding to the infrared image defect or the visible light image defect obtained with its corresponding preset threshold, and trigger an alarm if the category confidence of the defect category corresponding to the infrared image defect or the visible light image defect is greater than its corresponding preset threshold.
[0131] It should be understood that the device may also include a display interface for displaying defect detection results.
[0132] Embodiment 3,
[0133] This embodiment also provides an electronic device, including:
[0134] at least one processor; and
[0135] A memory storing instructions, which, when executed by the at least one processor, enables the at least one processor to execute the defect detection method for unmanned aerial vehicle inspection of photovoltaic arrays as described above.
[0136] In this embodiment, the electronic device may include, but is not limited to: personal computers, server computers, workstations, desktop computers, laptop computers, notebook computers, mobile computing devices, smart phones, tablet computers, cellular phones, personal digital assistants (PDAs), handheld devices, messaging devices, wearable computing devices, consumer electronic devices, and the like.
[0137] Embodiment 4,
[0138] This embodiment also provides a machine-readable storage medium storing executable instructions, which, when executed, enable the machine to execute the defect detection method for unmanned aerial vehicle inspection of photovoltaic arrays as described above.
[0139] Specifically, a system or device equipped with a readable storage medium can be provided, on which software program codes that implement the functions of any of the above-mentioned embodiments are stored, and a computer or processor of the system or device can read and execute instructions stored in the readable storage medium.
[0140] In this case, the program code itself read from the machine-readable medium can realize the function of any one of the above embodiments, and thus the machine-readable code and the machine-readable storage medium storing the machine-readable code constitute part of this specification.
[0141] Examples of readable storage media include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, the program code may be downloaded from a server computer or a cloud via a communication network.
[0142] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0143] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0144] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0145] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0146] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solution of the present invention, and are not intended to limit the specific implementation methods of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the claims of the present invention shall be included in the protection scope of the claims of the present invention.
Claims
1. A defect detection method for photovoltaic arrays by drone inspection, characterized in that: The method comprises: S1, obtaining an infrared image and a visible light image of each photovoltaic array, and pairing the infrared image and the visible light image of each photovoltaic array to form a plurality of groups of images to be detected; S2, inputting the image to be detected into a photovoltaic array defect detection model, the photovoltaic array defect detection model includes a segmentation model and a defect detection model arranged in cascade, the image to be detected is segmented by the segmentation model to generate a number of photovoltaic modules, the photovoltaic modules are input into the defect detection model after intermediate processing, and the corresponding defect detection results are output, the defect detection results include defect categories and category confidence; The segmentation model includes an infrared image segmentation model for segmenting infrared images and a visible light image segmentation model for segmenting visible light images, the defect detection model includes an infrared image defect detection model for identifying infrared images and a visible light image defect detection model for identifying visible light images, and the infrared image segmentation model and the infrared image defect detection model are cascaded, and the visible light image segmentation model and the visible light image defect detection model are cascaded; S3. Post-processing the defect detection result of the image to be detected to determine whether it is a defective image. If so, further determine whether it is an infrared image defect or a visible light image defect.
2. The defect detection method for photovoltaic arrays inspected by drones according to claim 1 is characterized in that: In step S2, the segmentation model adopts the Yolov8l-seg-640 model, and the defect detection model adopts the Yolov8l-dec-640 model optimized by the coordinate attention mechanism, that is, a three-layer attention module CA is inserted into the Yolov8l-dec-640 model, and the three-layer attention module CA serves as the next layer structure of the second C2F layer, the next layer structure of the third C2F layer, and the next layer structure of the fourth C2F layer in the Yolov8l-dec-640 model structure, respectively.
3. The defect detection method for photovoltaic arrays inspected by drones according to claim 1 is characterized in that: The step S2 further includes: constructing a segmentation data set and a defect detection data set based on the training data set to train the segmentation model and the defect detection model respectively, wherein the training data set includes training samples formed by pairing an infrared image and a visible light image of each photovoltaic array; The segmentation dataset and the defect detection dataset are constructed based on the training dataset, specifically including: Selecting some training samples from the training data set, and manually labeling all complete photovoltaic components in the entire image of each training sample to form segmentation data, thereby constructing a segmentation data set for training a segmentation model; Cutting the training samples that have been labeled with all the complete photovoltaic modules, and cutting the remaining unlabeled training samples in the training data set using the trained segmentation model to form a number of photovoltaic module samples; The defect locations in all photovoltaic module samples are marked by manual labeling to form defect detection data, thereby constructing a defect detection dataset for training defect detection models.
4. The defect detection method for photovoltaic arrays inspected by drones according to claim 1 is characterized in that: In the step S2, the photovoltaic module is subjected to intermediate processing, which specifically includes: Performing morphological operations of corrosion and expansion on the photovoltaic modules segmented in the image to be detected, so that the segmented edges are connected to form a number of connected domains; wherein, for the photovoltaic modules in the visible light image, a 23*23 structural element is used for expansion; and for the photovoltaic modules in the infrared image, a 5*5 structural element is used for expansion; Determine the number of connected domains formed in the image to be detected after expansion: If the number of connected domains is one, the connected domain is determined as the photovoltaic string to be tested; If the number of connected domains is greater than one, the distance between the centroid of each connected domain and the center point of the image to be detected is calculated, and the connected domain where the centroid of the minimum distance is located is determined as the photovoltaic string to be tested; The step of calculating the distance between the centroid of each connected domain and the center point of the image to be detected to determine the photovoltaic string to be detected specifically includes: Use opencv software to obtain the height H and width W of the image to be detected, and use (W / 2, H / 2) as the center point coordinates of the image to be detected; The number of all connected domains in the image to be detected and the coordinate information of each connected domain are obtained by the cv2.connectedComponentsWithStats() method, and the height and width of each connected domain are determined according to the coordinate information of each connected domain, so as to calculate the centroid coordinates of each connected domain; The distance between the coordinates of the centroid of each connected domain and the coordinates of the center point of the image to be detected is calculated by the Pythagorean theorem, and the connected domain where the centroid of the minimum distance is located is determined as the photovoltaic string to be detected.
5. The defect detection method for photovoltaic arrays inspected by drones according to claim 4 is characterized in that: In the step S2, the photovoltaic module is subjected to intermediate processing, which specifically includes: The highest point and the lowest point of the photovoltaic string to be tested are obtained according to the bounding box of the connected domain, and half of the distance between the highest point and the lowest point is used as the height of each photovoltaic module in the photovoltaic string to be tested to determine the position of the upper and lower dividing lines of the photovoltaic string to be tested; The row number to which each photovoltaic module belongs is determined based on the relationship between the y-value of the ordinate of each photovoltaic module and the ordinate value of the dividing line, if: If the y-value of the ordinate of the photovoltaic module is less than the ordinate value of the dividing line, it belongs to the first row; If the y-value of the ordinate of the photovoltaic module is greater than the ordinate value of the dividing line, it belongs to the second row; The column number to which each photovoltaic module belongs is determined by the size of the x-coordinate of each photovoltaic module, and a set of serial numbers is formed as the serial number of the corresponding photovoltaic module; The first photovoltaic module in the photovoltaic string to be tested has a serial number (1,1).
6. The defect detection method for photovoltaic arrays inspected by drones according to claim 1 is characterized in that: The step S3 specifically includes: S31, judging whether the defect detection result is empty, if it is empty, the image to be detected is determined to be a defect-free image; if it is not empty, entering S32; S32, judging whether there is only one defect detection result and it is a normal result, if so, the image to be detected is determined to be a defect-free image; otherwise, entering S33; S33, sorting the defect detection results according to the size of the category confidence. If there are multiple results, among which the confidence of the normal result is the largest and greater than 0.8, the image to be detected is considered to be a defect-free image; otherwise, entering S34; S34. According to the defect categories contained in the infrared image and the visible light image, determine whether the defect detection result is an infrared image defect or a visible light image defect; wherein the defect categories contained in the infrared image include diode short circuit, shadow occlusion, vegetation occlusion and hot spot, and the defect categories contained in the visible light image include pollutant occlusion, shadow occlusion, vegetation occlusion and severe dust accumulation.
7. The defect detection method for photovoltaic arrays inspected by drones according to claim 6 is characterized in that: The method further includes: comparing the category confidence included in the defect detection result with a preset threshold corresponding thereto, and triggering an alarm if the category confidence in the defect detection result is greater than the preset threshold corresponding thereto; Among them, for the defect categories contained in the infrared image, the preset thresholds corresponding to the category confidence are: diode short circuit 0.7; shadow occlusion 0.6; vegetation occlusion 0.5; hot spot 0.6; for the defect categories contained in the visible light image, the preset thresholds corresponding to the category confidence are: pollutant occlusion 0.5; shadow occlusion 0.55; vegetation occlusion 0.5; severe dust accumulation 0.
4.
8. A device for realizing a defect detection method for a photovoltaic array by using an unmanned aerial vehicle for inspection, characterized in that: The device comprises: A data acquisition module is used to acquire the infrared image and visible light image of each photovoltaic array, and pair the infrared image and visible light image of each photovoltaic array to form a plurality of groups of images to be detected; An image segmentation and defect detection module is used to input the image to be detected into a photovoltaic array defect detection model, the photovoltaic array defect detection model includes a segmentation model and a defect detection model arranged in cascade, the image to be detected is segmented by the segmentation model to generate a number of photovoltaic modules, the photovoltaic modules are input into the defect detection model after intermediate processing, and the corresponding defect detection results are output, the defect detection results include defect categories and category confidence; The segmentation model includes an infrared image segmentation model for segmenting infrared images and a visible light image segmentation model for segmenting visible light images, the defect detection model includes an infrared image defect detection model for identifying infrared images and a visible light image defect detection model for identifying visible light images, and the infrared image segmentation model and the infrared image defect detection model are cascaded, and the visible light image segmentation model and the visible light image defect detection model are cascaded; a detection result post-processing module, used to post-process the defect detection result of the image to be detected to determine whether it is a defective image, and if so, further determine whether it is an infrared image defect or a visible light image defect; and The alarm module is used to compare the category confidence contained in the defect detection result with the corresponding preset threshold, and trigger an alarm if the category confidence in the defect detection result is greater than the corresponding preset threshold.
9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and A memory storing instructions, which, when executed by the at least one processor, enables the at least one processor to execute the defect detection method for unmanned aerial vehicle inspection of photovoltaic arrays as described in any one of claims 1 to 7.
10. A machine-readable storage medium, characterized in that: The machine-readable storage medium stores executable instructions, and when the instructions are executed, the machine executes the defect detection method for unmanned aerial vehicle inspection of photovoltaic arrays as described in any one of claims 1 to 7.
Citation Information
Patent Citations
Photovoltaic module defect detection method and device, electronic equipment and storage medium
CN116309293A
Photovoltaic panel detection method, unmanned aerial vehicle and computer readable storage medium
CN116385421A
Cited By
Photovoltaic module defect diagnosis method, device and equipment
CN119338820A
Power station photovoltaic module defect detection method and device based on unsupervised learning
CN119648647A
Power plant photovoltaic module defect detection method and device based on unsupervised learning
CN119648647B
Defect detection method and device for photovoltaic module of power station, equipment and medium
CN119850526A
Multi-mode cooperative photovoltaic power station inspection method and system
CN119886491A
