Photovoltaic module damage intelligent detection method and system based on unmanned aerial vehicle image

Abstract

The invention provides a photovoltaic module damage intelligent detection method and system based on an unmanned aerial vehicle image, and relates to the technical field of unmanned aerial vehicle image.An unmanned aerial vehicle collects a multi-angle high-altitude image of a photovoltaic module array according to a predetermined path and records geographic position information; image splicing and overlap region correction are carried out by using the information to form a high-resolution panoramic image; in combination with a pre-constructed three-dimensional topological structure model, identifying and marking photovoltaic modules in the panorama, and generating a module distribution diagram; by analyzing the actual spectral response characteristics of each component, comparing standard data in a material science database, and screening suspected damage areas; applying an adaptive threshold algorithm and morphological operation to the areas to generate a feature map; and integrating real-time environmental factors and historical performance changes, extracting target features influencing the performance reduction of the photovoltaic module, predicting a damage state, and finally generating a detection report. According to the invention, the detection efficiency of the photovoltaic module is improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of drone images, and in particular, to an intelligent detection method and system for photovoltaic module damage based on drone images. Background Art

[0002] With the wide application of solar photovoltaic systems, higher requirements are put forward for the maintenance and monitoring of large-scale photovoltaic module arrays. To ensure the efficient and reliable operation of these systems, it is necessary to regularly check the status of photovoltaic modules to detect potential problems such as cracks, dirt, or aging. Traditional manual inspection methods are inefficient and costly, while the development of drone technology makes it possible to achieve fast, efficient, and low-cost monitoring of photovoltaic modules. Drones can automatically collect high-altitude images from multiple angles along a predetermined flight path and record geographical location information, thus providing an accurate data basis for subsequent image analysis.

[0003] Currently, the main methods for monitoring the status of photovoltaic modules include on-site manual inspection, fixed camera monitoring, and drone-based image acquisition. Among them, the drone-based method can already complete image acquisition and generate panoramic images, identify the positions of photovoltaic modules using 3D models, and locate problem areas by comparing the characteristic differences between intact and damaged modes through spectral analysis. In addition, some solutions have begun to attempt to combine machine learning algorithms for automated feature extraction and damage prediction.

[0004] Although the existing solutions have achieved automated monitoring of the status of photovoltaic modules to a certain extent, there are still several deficiencies. First, the traditional image stitching and correction processes are prone to introducing errors, affecting the accuracy of the final diagnosis results. Second, the existing spectral response characteristic analysis often lacks comprehensive consideration of real-time environmental factors and historical performance data, resulting in inaccurate damage prediction. Finally, although some solutions have begun to adopt machine learning methods, most have not fully utilized advanced technologies such as deep learning, transfer learning, and random forest, which limits the accuracy of feature extraction and the generalization ability of the prediction model. Therefore, developing a solution that integrates advanced image processing technology and artificial intelligence algorithms is crucial for improving the efficiency and accuracy of photovoltaic module monitoring. Summary of the Invention

[0005] The embodiments of the present application provide an intelligent detection method and system for photovoltaic module damage based on drone images to solve the problem of low detection efficiency of photovoltaic modules in the prior art.

[0006] In a first aspect, the embodiments of the present application provide an intelligent detection method for photovoltaic module damage based on drone images, including: Collect multi - angle high - altitude images of the photovoltaic module array collected by the drone on the predetermined flight path, and record the geographical location information corresponding to the multi - angle high - altitude images; Based on the geographical location information, splice and correct the overlapping areas of the multi - angle high - altitude images to obtain a high - resolution panoramic image; Using a pre - constructed three - dimensional topological structure model of photovoltaic modules, identify and locate the photovoltaic modules in the high - resolution panoramic image, and mark the photovoltaic modules to generate a component distribution map; Based on the component distribution map, analyze the actual spectral response characteristics of each photovoltaic module, obtain the reflection characteristic data of each photovoltaic module in a sound state and the spectral response characteristics under different damage modes from a pre - constructed materials science database as the standard spectral response characteristics, compare the actual spectral response characteristics and the standard spectral response characteristics, and screen out suspected damage areas; Use the adaptive threshold algorithm and morphological operations to process the suspected damage areas to generate suspected damage areas; Combining the suspected damage areas, the real - time environmental factors where the photovoltaic modules are located, and the historical performance change data, extract target features, identify the influence relationship between the target features and the performance degradation of the photovoltaic modules, and based on the target features and the influence relationship, predict the damage state of the photovoltaic modules to generate a detection report.

[0007] Optionally, using the adaptive threshold algorithm and morphological operations to process the suspected damage areas to generate suspected damage areas, including: According to the local gray - level distribution characteristics of each pixel point in the high - resolution panoramic image, process the suspected damage areas to obtain the optimal threshold corresponding to each pixel point; Based on the optimal threshold, divide the pixel points in the suspected damage areas into two categories: foreground and background to obtain classified pixel points. Based on the classified pixel points, generate a suspected damage area map, analyze the suspected damage area map to identify suspected damage patches, and generate suspected damage patch data; Based on the suspected damage patch data, determine the target structural element. Based on the target structural element, perform dilation, erosion operations, and morphological operations on the suspected damage area map to generate an optimized suspected damage area map to remove the noise in the suspected damage areas; Perform frequency - domain filtering on the optimized suspected damage area map to obtain a preliminary suspected damage feature map; Process the preliminary suspected damage feature map to generate suspected damage areas.

[0008] Optionally, according to the local gray - scale distribution characteristics of each pixel point in the high - resolution panoramic image, process the suspected damage area to obtain the optimal threshold corresponding to each pixel point, including: Perform smoothing processing on the local neighborhood defined for each pixel point in the high - resolution panoramic image to obtain a local neighborhood image; Analyze the gray - scale value distribution of all pixel points in the local neighborhood image to establish a local gray - scale distribution model, and apply histogram equalization technology to process the local neighborhood image to obtain a local image; Based on the local image and the local gray - scale distribution model, construct a composite objective function based on the structural similarity index and edge regularization. Based on the composite objective function, evaluate the segmentation effect of the suspected damage area under different thresholds to obtain a segmentation evaluation result. Combine the segmentation evaluation result and the cross - validation mechanism to optimize the composite objective function to obtain an optimized objective function; Define an initial candidate threshold, combine the optimized objective function to optimize the initial candidate threshold to obtain an optimized threshold, and use the optimized threshold as the optimal threshold of the local gray - scale distribution characteristics.

[0009] Optionally, the step of constructing a composite objective function based on the structural similarity index and edge regularization based on the local image and the local gray - scale distribution model, evaluating the segmentation effect of the suspected damage area under different thresholds based on the composite objective function to obtain a segmentation evaluation result, and combining the segmentation evaluation result and the cross - validation mechanism to optimize the composite objective function to obtain an optimized objective function includes: Based on the local image and the local gray - scale distribution model, evaluate the segmentation effect of the suspected damage area, perform edge detection on the local image to obtain edge information, and construct a composite objective function based on the edge information combined with an edge regularization term; Based on the composite objective function, perform segmentation processing on the suspected damage area to obtain a segmentation result, and use the segmentation result to evaluate the segmentation effect of the suspected damage area to generate a set of segmentation evaluation results; Divide the set of segmentation evaluation results into multiple subsets. According to the subsets, use a part of the subsets as a test set and the remaining subsets as a training set. Use the training set and the test set to perform multiple iterative evaluations on the segmentation effect of the suspected damage area to obtain an optimized objective function.

[0010] Optionally, based on the component distribution map, analyze the actual spectral response characteristics corresponding to each photovoltaic component, obtain the reflection characteristic data of each photovoltaic component in a sound state and the spectral response characteristics under different damage modes from a pre-constructed materials science database as the standard spectral response characteristics, compare the actual spectral response characteristics with the standard spectral response characteristics, and screen out suspected damaged areas, including: Based on the component distribution map, analyze the actual spectral response characteristics corresponding to each photovoltaic component to obtain characteristic information; Based on the characteristic information, partition the high-resolution panoramic image to obtain photovoltaic component areas, optimize the boundaries of the photovoltaic component areas to obtain a segmentation result, and analyze the segmentation result to generate a photovoltaic component area map; Based on the photovoltaic component area map, construct a deep belief network, adjust the hyperparameters of the deep belief network to obtain an optimized deep belief network, and use the optimized deep belief network to perform data analysis processing on the reflection characteristics of the photovoltaic component area map to generate the reflection characteristic data corresponding to the materials of each photovoltaic component in the photovoltaic component area; Extract key spectral features from the spectral data in the reflection characteristic data, perform anomaly classification processing on the key spectral features to obtain classified key spectral features, use the classified key spectral features to determine the spectral response features to be optimized, and dynamically adjust the spectral response features to be optimized based on the actual environmental illumination conditions to obtain the standard spectral response characteristics; Compare the actual spectral response characteristics with the standard spectral response characteristics to generate spectral analysis parameters, and based on the spectral analysis parameters, perform spectral analysis on each photovoltaic component area to screen out suspected damaged areas.

[0011] Optionally, the step of based on the photovoltaic component area map, constructing a deep belief network, adjusting the hyperparameters of the deep belief network to obtain an optimized deep belief network, and using the optimized deep belief network to perform data analysis processing on the reflection characteristics of the photovoltaic component area map to generate the reflection characteristic data corresponding to the materials of each photovoltaic component in the photovoltaic component area includes: Construct a deep belief network based on the data of the photovoltaic component area map; Combine the Gaussian process surrogate model and the acquisition function to search the hyperparameter space of the deep belief network to obtain an optimal hyperparameter combination, perform performance evaluation processing on the optimal hyperparameter combination to obtain an evaluation result, and select the best hyperparameter configuration based on the evaluation result to optimize the deep belief network and generate an optimized deep belief network; Based on the optimized deep belief network, data analysis and processing of the reflection characteristics are performed on the photovoltaic module area map to generate reflection behavior data. According to the standard reflection behavior data of existing standard samples, the reflection behavior data is corrected and verified to obtain the reflection characteristic data corresponding to the materials of each photovoltaic module in the photovoltaic module area.

[0012] Optionally, the method of using the pre-constructed three-dimensional topological structure model of the photovoltaic module to identify and locate the photovoltaic modules in the high-resolution panoramic image and mark the photovoltaic modules to generate a component distribution map includes: Using the pre-constructed three-dimensional topological structure model of the photovoltaic module to perform automatic identification and positioning processing on the photovoltaic modules in the high-resolution panoramic image to generate an identification result; Performing marking processing on the identification result to establish the topological connection relationship between the photovoltaic modules and generate a photovoltaic module distribution network; Based on the photovoltaic module distribution network, a component distribution map is generated, where the component distribution map records the positions and mutual relationships of all photovoltaic modules.

[0013] In a second aspect, an embodiment of the present application provides an intelligent detection system for photovoltaic module damage based on drone images, including: A collection module, configured to collect multi-angle high-altitude images of a photovoltaic module array collected by a drone on a predetermined flight path and record the geographical location information corresponding to the multi-angle high-altitude images; A correction module, configured to perform stitching and overlapping area correction on the multi-angle high-altitude images based on the geographical location information to obtain a high-resolution panoramic image; An identification module, configured to use the pre-constructed three-dimensional topological structure model of the photovoltaic module to identify and locate the photovoltaic modules in the high-resolution panoramic image and mark the photovoltaic modules to generate a component distribution map; An analysis module, configured to analyze the actual spectral response characteristics corresponding to each photovoltaic module based on the component distribution map, obtain the reflection characteristic data of each photovoltaic module in a sound state and the spectral response characteristics under different damage modes from a pre-constructed material science database as standard spectral response characteristics, compare the actual spectral response characteristics with the standard spectral response characteristics, and screen out suspected damage areas; A generation module, configured to combine the suspected damage areas, real-time environmental factors where the photovoltaic modules are located, and historical performance change data, extract target features, identify the influence relationship between the target features and the performance degradation of the photovoltaic modules, and predict the damage state of the photovoltaic modules based on the target features and the influence relationship to generate a detection report.

[0014] In a third aspect, an embodiment of the present application provides a computing device, including a processor and a memory. A computer program is stored in the memory, and the processor is configured to run the computer program to execute any one of the methods for intelligent detection of photovoltaic module damage based on UAV images in the first aspect.

[0015] In a fourth aspect, an embodiment of the present application provides a computer storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, they implement any one of the methods for intelligent detection of photovoltaic module damage based on UAV images in the first aspect.

[0016] In an embodiment of the present application, multi-angle high-altitude images of a photovoltaic module array collected by a UAV on a predetermined flight path are collected, and the geographical location information corresponding to the multi-angle high-altitude images is recorded; based on the geographical location information, the multi-angle high-altitude images are stitched and the overlapping areas are corrected to obtain a high-resolution panoramic image; a pre-constructed three-dimensional topological structure model of a photovoltaic module is used to identify and locate the photovoltaic modules in the high-resolution panoramic image, and the photovoltaic modules are marked to generate a component distribution map; based on the component distribution map, the actual spectral response characteristics corresponding to each photovoltaic module are analyzed, and the reflection characteristic data of each photovoltaic module in a sound state and the spectral response characteristics in different damage modes are obtained from a pre-constructed materials science database as standard spectral response characteristics, and the actual spectral response characteristics are compared with the standard spectral response characteristics to screen out suspected damage areas; the adaptive threshold algorithm and morphological operations are used to process the suspected damage areas to generate suspected damage areas; combined with the suspected damage areas, the real-time environmental factors where the photovoltaic modules are located and the historical performance change data, target features are extracted, and the influence relationship between the target features and the performance degradation of the photovoltaic modules is identified. Based on the target features and the influence relationship, the damage state of the photovoltaic modules is predicted to generate a detection report.

[0017] The technical solution of the present application has the following beneficial effects: The present application improves the detection efficiency of photovoltaic modules and reduces labor costs. Accurate damage location and prediction contribute to timely maintenance and extend the service life of the photovoltaic system. Reducing human errors through automated analysis improves the reliability of the diagnostic results. The combination of real-time monitoring and historical data analysis provides strong support for optimizing the operation of the photovoltaic system.

[0018] Furthermore, the embodiment of the present application also optimizes the processing process of the suspected damage area. By combining technical means such as the adaptive threshold algorithm based on Bayesian optimization, connected component analysis, adaptive selection of structural elements, and morphological operations, accurate segmentation and feature extraction of the suspected damage area in the high-resolution panoramic image of the photovoltaic module are achieved. In addition, through the application of the fast Fourier transform and the U-Net deep learning model, the quality of the suspected damage area is effectively improved, and the reliability of subsequent analysis is enhanced.

[0019] Through the above method, the accuracy and efficiency of identifying the suspected damage area in the monitoring process of the photovoltaic module are improved. First of all, the adaptive threshold algorithm optimized by Bayesian optimization can find the optimal threshold for each pixel point according to the local gray distribution characteristics, so as to more accurately separate the foreground and background and reduce misjudgment. Secondly, connected component analysis helps to identify independent suspected damage patches and generate more detailed suspected damage patch data, while the adaptive selection of structural elements ensures the effectiveness of morphological operations (such as dilation and erosion), and can better remove noise without damaging the real damage information. Finally, frequency domain filtering processing through the fast Fourier transform can eliminate unnecessary high-frequency noise, and the U-Net network model can automatically and intelligently extract the most representative suspected damage features from the preliminary image, and finally generate a high-quality suspected damage area, which not only improves the detection accuracy, but also provides a solid foundation for subsequent damage assessment.

[0020] These aspects or other aspects of the present application will be more clearly understood in the following description of the embodiments. Brief Description of the Drawings

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 It is a flowchart of an intelligent damage detection method for photovoltaic modules based on UAV images provided by an embodiment of the present application; Figure 2 It is a schematic structural diagram of an intelligent damage detection system for photovoltaic modules based on UAV images provided by an embodiment of the present application; Figure 3 It is a schematic structural diagram of a computing device provided by an embodiment of the present application. Detailed Embodiments

[0023] To enable those skilled in the art to better understand the solution of this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application.

[0024] In some processes described in the specification, claims and the above-mentioned drawings of this application, multiple operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear herein or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions such as "first" and "second" in this article are used to distinguish different messages, devices, modules, etc., do not represent a sequence, and do not limit that "first" and "second" are of different types.

[0025] The technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of this application.

[0026] Figure 1 The following is a flowchart of an intelligent detection method for photovoltaic module damage based on UAV images provided for the embodiments of this application. As Figure 1 shown, the method includes: Step 101: Collect multi-angle high-altitude images of the photovoltaic module array collected by the UAV on a predetermined flight path, and record the geographical location information corresponding to the multi-angle high-altitude images; In this step, the geographical location information refers to the position data obtained through positioning technologies such as GPS, including longitude, latitude, altitude, etc. These data can accurately describe the ground position corresponding to each image taken by the UAV. The multi-angle high-altitude images refer to images of the photovoltaic module array obtained from different perspectives, which are used to provide rich visual information.

[0027] In actual operation, the UAV flies according to a preset path and takes pictures of the photovoltaic module array at different heights and angles. Each time a picture is taken, the UAV records accurate geographical location information, which not only helps to ensure the geographical coherence between the images, but also provides the necessary spatial coordinate information for subsequent data processing.

[0028] For example, during the maintenance inspection of a large-scale solar power plant, an operator programmed one or more drones to fly automatically in a specific pattern to cover the entire plant area. The cameras on the drones captured images of the photovoltaic panels from multiple angles during flight, while recording the exact location of each picture. This data was then transmitted to a ground station to prepare for subsequent processing.

[0029] Step 102: Based on the geographical location information, stitch and correct the overlapping areas of the multi-angle high-altitude images to obtain a high-resolution panoramic image; In this step, stitching is the process of combining multiple images into a larger-range image; overlapping area correction is to eliminate the inconsistencies caused by changes in viewing angle or lighting between adjacent images to ensure the quality of the stitched image.

[0030] In actual operation, using image stitching software and combining geographical location information, images from different angles of the same photovoltaic module array are stitched into a complete high-resolution panoramic image. The software will automatically identify and adjust the overlapping parts between images to achieve a smooth transition while keeping the image details intact.

[0031] For example, continuing the above embodiment, after the drones complete the shooting task, engineers at the ground station use specialized image processing software to stitch all the uploaded images into a detailed panoramic image according to their geographical location information. This panoramic image can clearly show the layout of the photovoltaic panels in the entire power plant, laying a good foundation for the next analysis work.

[0032] Step 103: Using a pre-built three-dimensional topological structure model of photovoltaic modules, identify and locate the photovoltaic modules in the high-resolution panoramic image, and mark the photovoltaic modules to generate a component distribution map; In this step, the three-dimensional topological structure model is a mathematical model that represents the spatial relationships and connection methods of objects; deep learning algorithms are a class of artificial intelligence methods that can automatically extract features from a large amount of data; graph theory algorithms are used to solve problems in network structures, such as the shortest path between nodes.

[0033] In actual operation, based on a trained deep learning model, the system can automatically scan the panoramic image to find and label all the photovoltaic modules. Then, through graph theory algorithms, the system will mark each component and its connection method in the distribution map, thereby establishing the logical relationships between the components.

[0034] For example, based on the aforementioned panoramic image, engineers ran a deep learning model based on the U-Net architecture to identify each photovoltaic module and used graph theory algorithms to label them. The result is a detailed component distribution map that intuitively shows the specific locations of all the photovoltaic panels in the power plant and their interconnections.

[0035] Step 104: Based on the component distribution map, analyze the actual spectral response characteristics corresponding to each photovoltaic component, obtain the reflection characteristic data of each photovoltaic component in a sound state and the spectral response characteristics under different damage modes from a pre-constructed materials science database as the standard spectral response characteristics, compare the actual spectral response characteristics with the standard spectral response characteristics, and screen out the suspected damaged areas; In this step, the spectral response characteristic refers to the degree of response of a material to light of different wavelengths; the reflection characteristic data in a sound state is the ideal spectral response of the material when it is not damaged; the spectral response characteristics under damage modes refer to the spectral performance of the material when it is damaged in different forms.

[0036] In actual operation, first obtain the actual spectral response characteristics of each photovoltaic component, then query the materials science database to obtain the spectral response standards of the corresponding components under non-damaged conditions and various possible damage conditions. Finally, compare the two to find any deviations and thus lock in potential problem areas.

[0037] For example, engineers conducted spectral analysis for each photovoltaic component according to the component distribution map, and with reference to the standard spectral response characteristics in the materials science database, successfully identified several areas with abnormal spectral responses. These areas were marked as suspected damaged areas, providing key targets for further in-depth analysis.

[0038] Step 105: Process the suspected damaged areas using an adaptive threshold algorithm and morphological operations to generate suspected damaged areas; In this step, the adaptive threshold algorithm is a method for dynamically adjusting the image segmentation threshold, which can optimize the segmentation effect according to the local characteristics of the image; morphological operations are a series of techniques for changing the shape of an image, such as dilation and erosion, which help to clean up image noise and enhance target features.

[0039] In actual operation, segment the suspected damaged areas using an adaptive threshold algorithm optimized by Bayesian optimization, and then use morphological operations such as dilation and erosion to remove unnecessary noise, finally generating optimized suspected damaged areas to provide key evidence for damage assessment.

[0040] For example, engineers further analyzed the previously marked suspected damaged areas, applied advanced image processing techniques, including adaptive threshold segmentation and morphological operations, to generate clearer suspected damaged areas. This enables technicians to more easily distinguish where real damage may exist.

[0041] Step 106: Combine the suspected damage area, the real-time environmental factors where the photovoltaic module is located, and the historical performance change data, extract target features, identify the influence relationship between the target features and the performance degradation of the photovoltaic module, and predict the damage state of the photovoltaic module based on the target features and the influence relationship to generate a detection report.

[0042] In this step, transfer learning refers to applying the knowledge learned in one field to another related field; a convolutional neural network is an artificial intelligence model that is good at processing image data; random forest is an ensemble learning algorithm that performs classification or regression prediction by combining the results of multiple decision trees.

[0043] In actual operation, integrate all available information, and model and predict the state of the photovoltaic module through machine learning algorithms. In particular, use transfer learning to improve the generalization ability of the model, and then use the random forest algorithm for classification prediction to output the prediction results and recommended measures.

[0044] For example, finally, the engineers integrated the suspected damage area, the current environmental conditions, and the past performance data, and used an enhanced convolutional neural network and a random forest algorithm to predict the health status of the photovoltaic module. Based on the prediction results, a detailed detection report was compiled, indicating the specific components that need maintenance and proposing corresponding repair suggestions.

[0045] Through the above six steps, the accuracy and efficiency of photovoltaic module damage detection are effectively improved. From the accurate acquisition of multi-angle high-altitude images, to the efficient analysis achieved through deep learning and image processing technologies, and then to the damage prediction based on machine learning, each step is improving the intelligent level of photovoltaic power station operation and maintenance management. This solution can not only timely detect and locate possible problems, but also predict future damage trends through scientific analysis methods, greatly ensuring the safe and stable operation of the photovoltaic power station.

[0046] To solve the possible noise interference in the suspected damage area and improve the accuracy of damage patch recognition, in some embodiments, in step 105, the adaptive threshold algorithm and morphological operations are used to process the suspected damage area to generate a suspected damage area, including: According to the local gray-scale distribution characteristics of each pixel point in the high-resolution panoramic image, process the suspected damage area to obtain the optimal threshold corresponding to each pixel point; based on the optimal threshold, divide the pixel points in the suspected damage area into two categories: foreground and background, to obtain the classified pixel points, and based on the classified pixel points, generate a suspected damage area map, analyze the suspected damage area map to identify suspected damage patches, and generate suspected damage patch data; based on the suspected damage patch data, determine the target structural element, and based on the target structural element, perform dilation, erosion operations and morphological operations on the suspected damage area map to generate an optimized suspected damage area map to remove the noise in the suspected damage area; perform frequency-domain filtering on the optimized suspected damage area map to obtain a preliminary suspected damage feature map; process the preliminary suspected damage feature map to generate a suspected damage area.

[0047] In this embodiment, the local gray-scale distribution characteristic refers to the statistical attribute of pixel values within a certain range around each pixel point, such as average brightness or contrast. The adaptive threshold algorithm based on Bayesian optimization is a method for dynamically adjusting the segmentation threshold, which can automatically adjust the threshold according to the image content to make the segmentation result more suitable for the actual scene. The division of foreground and background categories is part of the image binarization process, which is used to distinguish important information and other irrelevant parts in the image. Connected component analysis is an image processing method used to find and label pixel groups that are connected to each other in the image, that is, connected regions, which helps to identify independent damage patches. The structural element adaptive selection algorithm is used to determine the template shape and size used in morphological operations to adapt to damage patches of different shapes and sizes. Morphological operations can enhance specific shape features in the image while reducing noise. The fast Fourier transform is a technique for converting a spatial-domain signal to the frequency domain, which can be used to filter out unwanted frequency components to clean up the image. U-Net is a convolutional neural network architecture, especially suitable for image segmentation tasks, and can accurately locate and describe objects in the image.

[0048] In the embodiments of the present application, first, for the suspected damage area, by analyzing the local gray-scale distribution characteristics of each pixel point, an adaptive threshold algorithm optimized by Bayesian is used to calculate an optimal segmentation threshold for each pixel. Next, based on these thresholds, the pixel points in the image are classified into two categories: foreground and background, and a suspected damage area map is constructed on this basis. After that, the suspected damage patches are identified through a connected component analysis algorithm, and information such as their positions and shapes is recorded. To remove the noise in the image, the most suitable structural element is selected for morphological operations to make the image clearer. Then, the fast Fourier transform technology is used to perform frequency-domain filtering on the image to eliminate high-frequency noise, and a preliminary map of suspected damage features is obtained. Finally, the U-Net network model is used to further process the preliminary map to generate a more refined and accurate suspected damage area.

[0049] The following is a specific example: During the daily maintenance inspection of a large solar power station, technicians found that there might be potential problems with some photovoltaic modules. According to the above method, first, high-resolution panoramic images of the power station were taken by drones. Subsequently, in the laboratory, engineers used an adaptive threshold algorithm based on Bayesian optimization to process the suspected damage area, obtained the optimal threshold for each pixel point, and classified the pixel points in the image into foreground and background accordingly. Further, multiple suspected damage patches were found through connected component analysis, and their data were recorded. To ensure that the identified suspected damage areas are as real as possible, engineers used morphological operations to remove noise and performed fast Fourier transform to remove the noise in the frequency domain. Finally, engineers used the U-Net network model to carefully analyze the processed image and generated the final suspected damage area. This feature map not only helped technicians accurately locate all the positions of suspected damage but also provided valuable guidance information for subsequent maintenance work.

[0050] To further improve the accuracy of suspected damage area detection and reduce the influence of image noise on the segmentation result, in some embodiments, when processing the suspected damage area according to the local gray-scale distribution characteristics of each pixel point in the high-resolution panoramic image in step 105 to obtain the optimal threshold corresponding to each pixel point, it further includes: Smoothing the local neighborhood defined for each pixel point in the high-resolution panoramic image to obtain a local neighborhood image; analyzing the gray value distribution of all pixel points in the local neighborhood image to establish a local gray distribution model, applying histogram equalization technology to process the local neighborhood image to obtain a local image; based on the local image and the local gray distribution model, constructing a composite objective function based on the structural similarity index and edge regularization, based on the composite objective function, evaluating the segmentation effect of the suspected damage area under different thresholds to obtain a segmentation evaluation result, combining the segmentation evaluation result and the cross-validation mechanism to optimize the composite objective function to obtain an optimized objective function; defining an initial candidate threshold, combining the optimized objective function to optimize the initial candidate threshold to obtain an optimized threshold, and using the optimized threshold as the optimal threshold of the local gray distribution characteristics. Optionally, the constructing a composite objective function based on the structural similarity index and edge regularization based on the local image and the local gray distribution model, based on the composite objective function, evaluating the segmentation effect of the suspected damage area under different thresholds to obtain a segmentation evaluation result, combining the segmentation evaluation result and the cross-validation mechanism to optimize the composite objective function to obtain an optimized objective function includes: evaluating the segmentation effect of the suspected damage area based on the local image and the local gray distribution model, and performing edge detection on the local image to obtain edge information, constructing a composite objective function based on the edge information combined with an edge regularization term; based on the composite objective function, performing segmentation processing on the suspected damage area to obtain a segmentation result, using the segmentation result to evaluate the segmentation effect of the suspected damage area to generate a segmentation evaluation result set; dividing the segmentation evaluation result set into multiple subsets, according to the subsets, taking a part of the subsets as a test set, and the remaining subsets as a training set, using the training set and the test set to perform multiple iterative evaluations on the segmentation effect of the suspected damage area to obtain an optimized objective function.

[0051] In this embodiment, the weighted average filter is a linear filtering method that smooths an image by applying different weights to each pixel point in the image and its surrounding pixel points, which can effectively reduce random noise while protecting important edge features. The local gray-level distribution model is a statistics-based method used to describe the distribution of gray-level values of pixel points within a specific region, which helps to understand the image content. Histogram equalization is a common image enhancement technique aimed at expanding the gray-level range of an image and increasing visual contrast. The structural similarity index is a metric used to measure the similarity between two images, which takes into account information from three aspects: brightness, contrast, and structure. The edge regularization term is a constraint condition for the image edges to ensure that edge features can be accurately preserved during the optimization process. The composite objective function combines multiple evaluation criteria into a single objective for the optimization process. The segmentation evaluation result set is a record of the segmentation effects at different thresholds, which is used to guide subsequent optimization work. The cross-validation mechanism is a method for evaluating the performance of a model by dividing the dataset into several subsets, training and testing the model repeatedly to find the optimal parameter configuration.

[0052] In the embodiment of the present application, first, the weighted average filter is applied to each pixel point in the high-resolution panoramic image to smooth the image while protecting edge details and generate a local neighborhood image. Then, by analyzing the gray-level value distribution of pixel points in these local neighborhood images, a local gray-level distribution model is established, which is crucial for subsequent threshold calculation. Subsequently, the histogram equalization technique is applied to improve the image quality and make the image clearer. Next, based on the improved local image and the previously established local gray-level distribution model, a composite objective function that combines the structural similarity index and edge regularization is constructed to evaluate the segmentation effects under different threshold settings. To ensure the robustness and generalization ability of the segmentation effect, a cross-validation mechanism is introduced to continuously iterate and optimize the composite objective function until a satisfactory result is obtained. Finally, under the guidance of the composite objective function, the initially set threshold is fine-tuned through Bayesian optimization technology to finally find the optimal threshold that best suits the local gray-level distribution characteristics of the current image.

[0053] The following is a specific example: In an automobile manufacturing factory, engineers need to regularly inspect the body surface for defects such as scratches or dents. An automated detection system is used, which is equipped with high-resolution cameras that can quickly capture panoramic images of the body on the production line. However, due to changes in lighting and different shooting angles, the original images may contain a large amount of noise, which poses challenges to automatic detection. According to the above method, engineers first performed weighted average filtering on the collected panoramic images, effectively reducing image noise while maintaining the clarity of key features such as scratches and dents. Then, a local gray distribution model was established, and the image contrast was enhanced through histogram equalization to make potential damages more obvious. Next, engineers constructed a composite objective function based on the structural similarity index and edge regularization to evaluate the segmentation effects at different thresholds. To ensure that the selected threshold is optimal, a cross-validation mechanism was adopted, and the composite objective function was optimized through multiple iterations, and finally the most suitable threshold was found. This process not only improved the accuracy of the detection system but also greatly shortened the detection time and enhanced the overall efficiency of the production line.

[0054] To solve the problem that potential damages of photovoltaic modules are difficult to detect by conventional means, in some embodiments, in step 104, based on the component distribution map, the actual spectral response characteristics corresponding to each photovoltaic module are analyzed, and the reflection characteristic data of each photovoltaic module in a sound state and the spectral response characteristics under different damage modes are obtained from a pre-constructed materials science database as standard spectral response characteristics, and the actual spectral response characteristics are compared with the standard spectral response characteristics to screen out suspected damage areas, including: Based on the component distribution map, analyze the actual spectral response characteristics corresponding to each photovoltaic component to obtain characteristic information; based on the characteristic information, perform zoning processing on the high-resolution panoramic image to obtain the photovoltaic component area, and optimize the boundary of the photovoltaic component area to obtain a segmentation result, and analyze the segmentation result to generate a photovoltaic component area map; based on the photovoltaic component area map, construct a deep belief network, and adjust the hyperparameters of the deep belief network to obtain an optimized deep belief network, and use the optimized deep belief network to perform data analysis processing on the reflection characteristics of the photovoltaic component area map to generate the reflection characteristic data corresponding to the materials of the photovoltaic components in the photovoltaic component area; extract key spectral features from the spectral data in the reflection characteristic data, and perform anomaly classification processing on the key spectral features to obtain classified key spectral features, use the classified key spectral features to determine the spectral response characteristics to be optimized, and dynamically adjust the spectral response characteristics to be optimized based on the actual environmental illumination conditions to obtain standard spectral response characteristics; compare the actual spectral response characteristics with the standard spectral response characteristics to generate spectral analysis parameters, and based on the spectral analysis parameters, perform spectral analysis on each photovoltaic component area to screen out suspected damaged areas. Optionally, the step of based on the photovoltaic component area map, constructing a deep belief network, and adjusting the hyperparameters of the deep belief network to obtain an optimized deep belief network, and using the optimized deep belief network to perform data analysis processing on the reflection characteristics of the photovoltaic component area map to generate the reflection characteristic data corresponding to the materials of the photovoltaic components in the photovoltaic component area includes: constructing a deep belief network based on the data of the photovoltaic component area map; combining a Gaussian process surrogate model and an acquisition function to search the hyperparameter space of the deep belief network to obtain an optimal hyperparameter combination, perform performance evaluation processing on the optimal hyperparameter combination to obtain an evaluation result, and select the best hyperparameter configuration based on the evaluation result to optimize the deep belief network to generate an optimized deep belief network; based on the optimized deep belief network, perform data analysis processing on the reflection characteristics of the photovoltaic component area map to generate reflection behavior data, and perform calibration and verification processing on the reflection behavior data according to the standard reflection behavior data of existing standard samples to obtain the reflection characteristic data corresponding to the materials of the photovoltaic components in the photovoltaic component area.

[0055] In this embodiment, the spectral response characteristic refers to the response efficiency of a photovoltaic module to light of different wavelengths, which is one of the important indicators for evaluating the performance of a photovoltaic module. The materials science database is a pre-constructed resource library that contains spectral response data of various materials in their intact state and different damage modes. These data are used as a standard reference for comparing the spectral response characteristics of the actually measured photovoltaic modules to identify potential problems. The superpixel segmentation algorithm is an image segmentation method that decomposes an image into several small regions with similar attributes, namely superpixels, thus simplifying the image processing task. The maximum flow minimum cut algorithm is a graph theory algorithm used to find the optimal segmentation boundary in an image, maximizing the difference between adjacent regions. Morphological operations and connected component analysis are techniques used to clean and enhance the segmentation results, ensuring that the boundaries of each photovoltaic module are clear and continuous. The deep belief network is a deep neural network structure that can learn complex non-linear mapping relationships, while Bayesian optimization is used to automatically tune the hyperparameters in the network to improve the prediction accuracy. Principal component analysis is a statistical method that can reduce the data dimension while retaining the most important information. The support vector machine is a supervised learning model that is good at small sample, non-linear, and high-dimensional pattern recognition. The maximum margin principle is the core concept of the SVM, aiming to find a hyperplane to best separate data points of different classes. The genetic algorithm is a stochastic search algorithm that simulates natural selection and genetic mechanisms and is used to find the global optimal solution.

[0056] In the embodiment of this application, first, the actual spectral response characteristics of each photovoltaic module are analyzed according to the component distribution map to obtain characteristic information. Subsequently, the superpixel segmentation algorithm and the maximum flow minimum cut algorithm are applied to optimize the boundary to form a photovoltaic module area map. Then, a deep belief network is constructed and optimized to analyze the reflection characteristic data of the materials of each photovoltaic module. Then, the reflection characteristic data are processed by the principal component analysis dimensionality reduction technique and the support vector machine classifier to identify abnormal situations, determine the spectral response characteristics to be optimized, and adjust these characteristics according to the actual lighting conditions. Finally, the genetic algorithm is used to compare the actual spectral response characteristics with the standard characteristics to generate spectral analysis parameters, and based on this, the suspected damaged areas are screened out.

[0057] The following is a specific example: In a solar power station, technicians need to regularly check whether there are hidden damages in the photovoltaic modules. These damages may reduce the power generation efficiency but are not easily detectable by the naked eye. According to the above method, a high-resolution panoramic image of the power station was first taken by a drone, and the actual spectral response characteristics of each photovoltaic module were analyzed based on the module distribution map. Next, the engineers used the superpixel segmentation algorithm to partition the image and optimized the boundaries with the maximum flow minimum cut algorithm to obtain an accurate photovoltaic module area map. After that, a deep belief network was constructed, and its hyperparameters were finely tuned using Bayesian optimization to more accurately analyze the reflection characteristics of each photovoltaic module material. For further analysis, the engineers used principal component analysis technology to extract key spectral features and performed anomaly classification using a support vector machine classifier. Finally, the genetic algorithm was used to compare the actual spectral response characteristics with the standard characteristics, and the photovoltaic module areas with potential problems were successfully screened out, providing a clear direction for subsequent maintenance work.

[0058] To solve the accuracy problem of component identification and positioning in a photovoltaic power station and improve the detection efficiency and accuracy, in some embodiments, using the pre-constructed three-dimensional topological structure model of photovoltaic modules to identify and locate the photovoltaic modules in the high-resolution panoramic image and mark the photovoltaic modules to generate a component distribution map in step 103 includes: Using the pre-constructed three-dimensional topological structure model of photovoltaic modules to automatically identify and locate the photovoltaic modules in the high-resolution panoramic image to generate an identification result; performing a marking process on the identification result to establish the topological connection relationship between the photovoltaic modules to generate a photovoltaic module distribution network; generating a component distribution map based on the photovoltaic module distribution network, where the component distribution map records the positions and mutual relationships of all photovoltaic modules.

[0059] In this embodiment, the three-dimensional topological structure model of photovoltaic modules is a detailed digital model that includes the physical dimensions, shapes, and their relative positions in space of each module in the photovoltaic array. This model is not only used to assist deep learning algorithms to more accurately identify and locate photovoltaic modules in images but also serves as the basis for subsequent analysis. Deep learning algorithms are a type of machine learning technology, especially those artificial neural networks that contain multiple layers of nonlinear transformations, such as convolutional neural networks. They are good at automatically extracting features from large amounts of data and are particularly effective for image recognition tasks. Graph theory algorithms are a branch of mathematics that involves the study of graph structures and are used here to mark and define the connection methods between photovoltaic modules to help understand the topological relationships between components. The component distribution map is a visualization tool generated according to the above process, which clearly marks the positions of each photovoltaic module and their connections and is crucial for the management and maintenance of photovoltaic power stations.

[0060] In the embodiments of the present application, first, a three-dimensional topological structure model of photovoltaic modules and a deep learning algorithm are used to process high-resolution panoramic images obtained by drones or other imaging devices, so as to realize the automatic identification and precise positioning of photovoltaic modules. Then, the identified modules are connected by graph theory algorithms to construct a network reflecting the actual connection relationship between the modules. Finally, a module distribution map is generated based on this network. This map not only shows the specific positions of the modules but also reveals how they are connected together, which is very useful for the monitoring and fault troubleshooting of photovoltaic power stations.

[0061] The following is a specific example: In a large-scale ground photovoltaic power station, in order to ensure the efficient operation of the system, technicians need to regularly check the status of photovoltaic modules. First, a three-dimensional topological structure model containing detailed information of all photovoltaic modules in the power station is constructed. Then, during a routine inspection, the operator controls the drone to fly over the entire power station and captures a series of high-resolution panoramic images. The technicians input these images into a deep learning-based system, which can automatically identify and precisely position each photovoltaic module. Next, graph theory algorithms are used to label these modules and establish the topological connection relationship between the modules, forming a complete photovoltaic module distribution network. Finally, based on this network, a detailed module distribution map is generated. This map not only helps with daily management but also can quickly locate possible problem areas to guide the work of the maintenance team. By this method, technicians can complete the inspection of the photovoltaic power station more quickly and accurately, thereby improving the overall operation and maintenance efficiency of the power station.

[0062] The present application considers that traditional threshold segmentation methods have limitations when dealing with photovoltaic module images, such as being sensitive to light changes and noise, and being unable to adapt to complex local gray distribution characteristics. In order to improve the accuracy and robustness of the identification of suspected damage areas, a method that can adaptively adjust the threshold is needed to cope with image changes under different environmental conditions. Therefore, a new alternative solution is proposed, which includes: Processing the suspected damage area according to the local gray distribution characteristics of each pixel point in the high-resolution panoramic image to obtain the optimal threshold corresponding to each pixel point, including: Smooth the local neighborhood defined for each pixel point in the high-resolution panoramic image, and obtain the local neighborhood image by the method of weighted average. The pixel value of the smoothed image at position (x, y) is obtained by calculating the weighted average of all pixel points in the neighborhood, where the weights are jointly determined by the spatial distance and the gray-scale similarity. The spatial distance weight indicates that the closer the pixel point is to the center point, the greater the weight; the gray-scale similarity weight indicates that the closer the gray-scale value of the pixel point is to the center point, the greater the weight. These two weights are calculated by the Gaussian function and are controlled by the spatial distance standard deviation and the gray-scale similarity standard deviation respectively.

[0063] The smoothed image reflects the gray-scale value at position (x, y) after weighted average filtering. The local neighborhood N(x, y) is a region defined with position (x, y) as the center, usually a rectangular window or a circular region. The spatial distance standard deviation controls the decay rate of the spatial distance weight, and the gray-scale similarity standard deviation controls the decay rate of the gray-scale similarity weight. These two parameters need to be determined according to experience or experiments.

[0064] Next, analyze the gray-scale value distribution of all pixel points in the local neighborhood image to establish a local gray-scale distribution model. Apply histogram equalization technology to process the local neighborhood image to obtain a local image. Histogram equalization enhances the contrast of the image by adjusting the gray-scale distribution of the image to make the gray-level distribution more uniform. The equalized gray-scale value is obtained by calculating the probability weighted average of the original gray levels, and then normalizing and mapping. Normalization ensures that local images of different sizes can be fairly compared, and mapping ensures that the equalized gray-scale values can fully utilize the entire gray-scale range.

[0065] Based on the local image and the local gray-scale distribution model, construct a composite objective function based on the structural similarity index and edge regularization. Based on the composite objective function, evaluate the segmentation effect of the suspected damage area under different thresholds to obtain a segmentation evaluation result. Combine the segmentation evaluation result and the cross-validation mechanism to optimize the composite objective function to obtain an optimized objective function.

[0066] Define an initial candidate threshold, and optimize the initial candidate threshold in combination with the optimized objective function to obtain an optimized threshold. Take the optimized threshold as the optimal threshold of the local gray-scale distribution characteristics. The goal of Bayesian optimization is to maximize the information gain, that is, the expected improvement. The information gain is evaluated by calculating the potential improvement amount of the new threshold relative to the current best value and combining the conditional probability density function. The conditional probability density function reflects the uncertainty and possibility of the model for the new threshold.

[0067] Based on the optimized parameter set, the initial threshold range is determined using the principle of maximum entropy. Based on the initial threshold range, Bayesian optimization is utilized to calculate the optimal threshold for the local gray-level distribution characteristics. The principle of maximum entropy takes into account the uncertainty of the local gray-level distribution when determining the threshold range. The optimal threshold is obtained by maximizing the entropy value. The larger the entropy value, the more uniform the distribution and the higher the information content.

[0068] The following is a specific embodiment: During the daily maintenance inspection of a large-scale ground-mounted photovoltaic power station, technicians need to regularly monitor the status of photovoltaic modules to detect potential damages in a timely manner. To address the problem that traditional threshold segmentation methods perform poorly under complex lighting conditions, the above-mentioned proposed solution is adopted. First, the technicians use drones to obtain high-resolution panoramic images of all photovoltaic modules in the power station. Next, these images are preprocessed. A weighted average filter is used to smooth the noise in the images while maintaining the integrity of important features such as edges. Subsequently, histogram equalization is performed on the smoothed images to enhance the image contrast, making subsequent analysis easier. On this basis, the technicians construct a composite objective function that comprehensively considers factors such as structural similarity and edge protection to evaluate the segmentation quality under different threshold settings. This objective function is continuously optimized through cross-validation to find the parameter configuration that is most suitable for the current image conditions. Then, a set of initial candidate thresholds is defined, and Bayesian optimization technology is applied to further refine these thresholds to find the best threshold that can maximize the information gain. Finally, the threshold range is reset based on the principle of maximum entropy, and Bayesian optimization is applied again to finally determine the optimal threshold for each local region. This method not only improves the accuracy of detecting suspected damage areas but also reduces the false alarm rate, greatly enhancing the efficiency and reliability of the operation and maintenance management of the photovoltaic power station.

[0069] Figure 2 The following is a schematic structural diagram of an intelligent detection system for photovoltaic module damage based on drone images provided by an embodiment of the present application, as Figure 2 shown. The system includes: A collection module 21, configured to collect multi-angle high-altitude images of a photovoltaic module array acquired by a drone on a predetermined flight path, and record the geographical location information corresponding to the multi-angle high-altitude images; A calibration module 22, configured to perform stitching and overlapping area calibration on the multi-angle high-altitude images based on the geographical location information to obtain a high-resolution panoramic image; An identification module 23, configured to identify and locate the photovoltaic modules in the high-resolution panoramic image by using a pre-constructed three-dimensional topological structure model of the photovoltaic modules, and mark the photovoltaic modules to generate a component distribution map; An analysis module 24, configured to analyze the actual spectral response characteristics corresponding to each photovoltaic module based on the component distribution map, obtain the reflection characteristic data of each photovoltaic module in a sound state and the spectral response characteristics under different damage modes from a pre-constructed materials science database as standard spectral response characteristics, compare the actual spectral response characteristics with the standard spectral response characteristics, and screen out suspected damaged areas; A generation module 25, configured to combine the suspected damaged areas, the real-time environmental factors where the photovoltaic modules are located, and the historical performance change data, extract target features, identify the influence relationship between the target features and the performance degradation of the photovoltaic modules, and predict the damage state of the photovoltaic modules based on the target features and the influence relationship to generate a detection report.

[0070] Figure 2 The intelligent detection system for photovoltaic module damage based on drone images described above can execute Figure 1 The intelligent detection method for photovoltaic module damage based on drone images described in the embodiments shown, the implementation principle and technical effects will not be elaborated. For the intelligent detection system for photovoltaic module damage based on drone images in the above embodiments, the specific ways for each module and unit to perform operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0071] In a possible design, Figure 2 The intelligent detection system for photovoltaic module damage based on drone images described in the embodiments shown can be implemented as a computing device, such as Figure 3 shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32.

[0072] The processing component 32 is used for the Figure 1 intelligent detection method for photovoltaic module damage based on drone images described in the above

[0073] Among them, the processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components for executing the above method.

[0074] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0075] Of course, the computing device may also necessarily include other components, such as an input / output interface, a display component, a communication component, etc.

[0076] The input / output interface provides an interface between the processing component and the peripheral interface module, and the above-mentioned peripheral interface module can be an output device, an input device, etc.

[0077] The communication component is configured to facilitate communication between the computing device and other devices in a wired or wireless manner, etc.

[0078] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. At this time, the computing device can refer to a cloud server, and the above-mentioned processing component, storage component, etc. can be basic server resources leased or purchased from a cloud computing platform.

[0079] The embodiment of the present application also provides a computer storage medium storing a computer program, and when the computer program is executed by a computer, it can implement the above Figure 1 intelligent detection method for photovoltaic module damage based on drone images shown in the above embodiments.

[0080] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0081] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0082] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. An intelligent detection method for photovoltaic module damage based on UAV images, characterized in that, Including: Collect multi-angle high-altitude images of a photovoltaic module array collected by a drone on a predetermined flight path, and record the geographical location information corresponding to the multi-angle high-altitude images; Based on the geographical location information, splice and correct the overlapping areas of the multi-angle high-altitude images to obtain a high-resolution panoramic image; Using a pre-constructed three-dimensional topological structure model of a photovoltaic module, identify and locate the photovoltaic modules in the high-resolution panoramic image, and mark the photovoltaic modules to generate a component distribution map; Based on the component distribution map, analyze the actual spectral response characteristics corresponding to each photovoltaic module, obtain the reflection characteristic data of each photovoltaic module in a sound state and the spectral response characteristics under different damage modes from a pre-constructed material science database as standard spectral response characteristics, compare the actual spectral response characteristics with the standard spectral response characteristics, and screen out suspected damage areas; Combining the suspected damage areas, the real-time environmental factors where the photovoltaic modules are located, and the historical performance change data, extract target features, identify the influence relationship between the target features and the performance degradation of the photovoltaic modules, and based on the target features and the influence relationship, predict the damage state of the photovoltaic modules to generate a detection report.

2. The method according to claim 1, wherein Processing the suspected damage areas using an adaptive threshold algorithm and morphological operations to generate suspected damage areas, including: Processing the suspected damage areas according to the local gray distribution characteristics of each pixel point in the high-resolution panoramic image to obtain the optimal threshold corresponding to each pixel point; Based on the optimal threshold, divide the pixel points in the suspected damage areas into two categories: foreground and background to obtain classified pixel points, and based on the classified pixel points, generate a suspected damage area map, analyze the suspected damage area map to identify suspected damage patches, and generate suspected damage patch data; Based on the suspected damage patch data, determine a target structural element, and based on the target structural element, perform dilation, erosion operations and morphological operations on the suspected damage area map to generate an optimized suspected damage area map to remove the noise in the suspected damage areas; Performing frequency domain filtering processing on the optimized suspected damage area map to obtain a preliminary suspected damage feature map; Processing the preliminary suspected damage feature map to generate a suspected damage area.

3. The method according to claim 2, wherein Processing the suspected damage areas according to the local gray distribution characteristics of each pixel point in the high-resolution panoramic image to obtain the optimal threshold corresponding to each pixel point, including: Smoothing the local neighborhood defined for each pixel point in the high-resolution panoramic image to obtain a local neighborhood image; Analyze the gray value distribution of all pixel points in the local neighborhood image to establish a local gray distribution model, and apply histogram equalization technology to process the local neighborhood image to obtain a local image; Based on the local image and the local gray distribution model, construct a composite objective function based on the structural similarity index and edge regularization. Based on the composite objective function, evaluate the segmentation effect of the suspected damage area under different thresholds to obtain a segmentation evaluation result. Combine the segmentation evaluation result and the cross-validation mechanism to optimize the composite objective function to obtain an optimized objective function; Define an initial candidate threshold, and optimize the initial candidate threshold in combination with the optimized objective function to obtain an optimized threshold, and use the optimized threshold as the optimal threshold of the local gray distribution characteristics.

4. The method according to claim 3, wherein The method of constructing a composite objective function based on the structural similarity index and edge regularization based on the local image and the local gray distribution model, evaluating the segmentation effect of the suspected damage area under different thresholds based on the composite objective function to obtain a segmentation evaluation result, and combining the segmentation evaluation result and the cross-validation mechanism to optimize the composite objective function to obtain an optimized objective function includes: Based on the local image and the local gray distribution model, evaluate the segmentation effect of the suspected damage area, and perform edge detection on the local image to obtain edge information. Based on the edge information and combined with the edge regularization term, construct a composite objective function; Based on the composite objective function, perform segmentation processing on the suspected damage area to obtain a segmentation result. Use the segmentation result to evaluate the segmentation effect of the suspected damage area to generate a set of segmentation evaluation results; Divide the set of segmentation evaluation results into multiple subsets. According to the subsets, use a part of the subsets as the test set and the remaining subsets as the training set. Use the training set and the test set to perform multiple iterative evaluations on the segmentation effect of the suspected damage area to obtain an optimized objective function.

5. The method according to claim 1, wherein The method of analyzing the actual spectral response characteristics corresponding to each photovoltaic module based on the component distribution map, obtaining the reflection characteristic data of each photovoltaic module in a sound state and the spectral response characteristics under different damage modes from a pre-constructed materials science database as the standard spectral response characteristics, and comparing the actual spectral response characteristics and the standard spectral response characteristics to screen out the suspected damage area includes: Based on the component distribution map, analyze the actual spectral response characteristics corresponding to each photovoltaic module to obtain characteristic information; Based on the characteristic information, perform zoning processing on the high-resolution panoramic image to obtain a photovoltaic module area, and optimize the boundary of the photovoltaic module area to obtain a segmentation result. Analyze the segmentation result to generate a photovoltaic module area map; Based on the photovoltaic module area map, construct a deep belief network, and adjust the hyperparameters of the deep belief network to obtain an optimized deep belief network. Use the optimized deep belief network to perform data analysis processing on the reflection characteristics of the photovoltaic module area map to generate the reflection characteristic data corresponding to the materials of each photovoltaic module in the photovoltaic module area; Extract key spectral features from the spectral data in the reflection characteristic data, perform anomaly classification processing on the key spectral features to obtain the classified key spectral features, use the classified key spectral features to determine the spectral response features to be optimized, and dynamically adjust the spectral response features to be optimized based on the actual environmental illumination conditions to obtain the standard spectral response characteristics; Compare the actual spectral response characteristics with the standard spectral response characteristics to generate spectral analysis parameters, and based on the spectral analysis parameters, perform spectral analysis on each photovoltaic module area to screen out suspected damaged areas.

6. The method according to claim 5, wherein Based on the photovoltaic module area map, construct a deep belief network, and adjust the hyperparameters of the deep belief network to obtain the optimized deep belief network. Use the optimized deep belief network to perform data analysis processing on the reflection characteristics of the photovoltaic module area map to generate the reflection characteristic data corresponding to the materials of each photovoltaic module in the photovoltaic module area, including: Based on the data of the photovoltaic module area map, construct a deep belief network; Combine the Gaussian process surrogate model and the acquisition function to search the hyperparameter space of the deep belief network to obtain the optimal hyperparameter combination, perform performance evaluation processing on the optimal hyperparameter combination to obtain the evaluation result, and select the best hyperparameter configuration based on the evaluation result to optimize the deep belief network and generate the optimized deep belief network; Based on the optimized deep belief network, perform data analysis processing on the reflection characteristics of the photovoltaic module area map to generate reflection behavior data, and perform calibration and verification processing on the reflection behavior data according to the standard reflection behavior data of existing standard samples to obtain the reflection characteristic data corresponding to the materials of each photovoltaic module in the photovoltaic module area.

7. The method according to claim 1, characterized in that, The use of a pre-constructed three-dimensional topological structure model of photovoltaic modules to identify and locate the photovoltaic modules in the high-resolution panoramic image and mark the photovoltaic modules to generate a component distribution map, including: Use a pre-constructed three-dimensional topological structure model of photovoltaic modules to perform automatic identification and location processing on the photovoltaic modules in the high-resolution panoramic image to generate an identification result; Perform marking processing on the identification result to establish the topological connection relationship between the photovoltaic modules and generate a photovoltaic module distribution network; Based on the photovoltaic module distribution network, generate a component distribution map, where the component distribution map records the positions and mutual relationships of all photovoltaic modules.

8. An intelligent detection system for photovoltaic module damage based on drone images, characterized in that, Including: A collection module for collecting multi-angle high-altitude images of a photovoltaic module array collected by a drone on a predetermined flight path and recording the geographical location information corresponding to the multi-angle high-altitude images; A calibration module for stitching and overlapping area calibration of the multi-angle high-altitude images based on the geographical location information to obtain a high-resolution panoramic image; An identification module for using a pre-constructed three-dimensional topological structure model of photovoltaic modules to identify and locate the photovoltaic modules in the high-resolution panoramic image and mark the photovoltaic modules to generate a component distribution map; An analysis module, configured to analyze the actual spectral response characteristics corresponding to each photovoltaic module based on the component distribution map, obtain the reflection characteristic data of each photovoltaic module in a sound state and the spectral response characteristics under different damage modes from a pre-constructed materials science database as standard spectral response characteristics, compare the actual spectral response characteristics with the standard spectral response characteristics, and screen out suspected damaged areas; A generation module, configured to combine the suspected damaged areas, the real-time environmental factors where the photovoltaic modules are located, and the historical performance change data, extract target features, identify the influence relationship between the target features and the performance degradation of the photovoltaic modules, and predict the damage state of the photovoltaic modules based on the target features and the influence relationship to generate a detection report.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for intelligent detection of photovoltaic module damage based on drone images according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that, A computer program is stored, and when the computer program is executed by a computer, it implements a method for intelligent detection of photovoltaic module damage based on drone images according to any one of claims 1 to 7.

Images