Photovoltaic panel defect detection method and system based on infrared camera carried by unmanned aerial vehicle

Abstract

The invention discloses a photovoltaic panel defect detection method and system based on an infrared camera carried by an unmanned aerial vehicle, and the method comprises the steps: collecting an EL image of a photovoltaic panel through the infrared camera carried by the unmanned aerial vehicle, building an image processing algorithm model, recognizing an abnormal region on the photovoltaic panel, and classifying faults; an artificial intelligence technology is utilized to accurately position an abnormal area on the photovoltaic panel, and finally an edge cloud cooperation strategy is adopted to realize lightweight deployment, so that the unmanned nondestructive detection and identification method capable of rapidly and accurately detecting the defects of the photovoltaic panel is provided.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition nondestructive testing, and mainly relates to a photovoltaic panel defect detection method and system based on an unmanned aerial vehicle equipped with an infrared camera. Background Art

[0002] With the popularity of solar power generation, the operation status detection and maintenance of photovoltaic panels have become increasingly important. Traditional detection methods often rely on manual visual inspection, which is not only time-consuming and labor-intensive, but also difficult to cover large-area photovoltaic arrays and sparse distributed photovoltaic arrays. In addition, detection work is almost impossible to carry out at night or in bad weather. As an important carrier for clean energy conversion, photovoltaic panels may encounter defects and failures caused by a series of external and internal factors during long-term operation, which will lead to a decrease in the energy conversion efficiency of photovoltaic panels, performance degradation and shortened life.

[0003] Photovoltaic panel defect detection is the cornerstone of ensuring the performance, safety and economy of photovoltaic systems. Various defects in photovoltaic panels, such as cracks in the cell, hot spot effect, poor electrical connection or deterioration of packaging materials, will significantly reduce the photoelectric conversion efficiency and directly affect the power generation and economic benefits of the system. At the same time, these defects also pose potential safety hazards. The hot spot effect may cause local overheating and cause fires. Hidden cracks in the cell are aggravated under extreme conditions, threatening personal and property safety. Failure of electrical connections increases the risk of short circuits or electric shocks. Regular inspection and maintenance are the key to preventing such problems.

[0004] In 2017, French scholar Eric Tomasela et al. proposed an advanced method for diagnosing the performance degradation of photovoltaic panels using current-voltage (IV) curve analysis. The core of this method is to accurately measure the IV curve of photovoltaic panels under specific light and temperature conditions, and then analyze a series of key photoelectric performance parameters, including maximum power point (MPP), open circuit voltage (Voc), short circuit current (Isc) and fill factor (FF). Among them, the maximum power point directly reflects the actual power generation capacity of the photovoltaic panel, while the fill factor comprehensively evaluates the output efficiency of the photovoltaic panel. By comparing the ideal parameters under standard conditions with the actual measured values, researchers can accurately identify whether the photovoltaic panel has performance degradation, such as defects such as cell cracks, hot spot effects or poor electrical connections. The advantage of this method is that it provides accurate quantitative evaluation and provides data support for photovoltaic system maintenance and performance optimization. However, the research of Tomasela et al. also revealed the limitations of the IV curve analysis method, that is, its implementation process is relatively cumbersome and requires independent testing of each photovoltaic panel, which means high time and labor costs in large-scale photovoltaic arrays.

[0005] To further improve the detection efficiency, Korean scholar Lee et al. proposed and verified a method of using ultrasonic detection technology to efficiently identify hidden defects inside photovoltaic panels. Ultrasonic detection technology makes full use of the good penetration ability and high reflection sensitivity of ultrasound, and can penetrate into the internal structure of photovoltaic panels to accurately locate imperceptible defects such as microcracks, delamination, and voids. Compared with traditional detection methods, ultrasonic detection can not only provide a non-invasive and comprehensive assessment, but also has higher detection sensitivity and resolution, so that even the slightest structural changes can be captured and analyzed. However, its disadvantage is that it requires high professional skills from operators.

[0006] In order to improve the automation of defect detection and reduce the reliance on professional skills, image-based automatic defect detection methods have been widely used in recent years with the development of computer vision and deep learning technologies. German scholar Volker Sel leneit et al. tried to automatically identify defects in images of photovoltaic panels by training deep learning models such as convolutional neural networks (CNN) in order to improve detection efficiency and accuracy. Therefore, how to fully and reasonably utilize computer vision and deep learning technologies has become a research direction for technicians in this field. Summary of the invention

[0007] The present invention is aimed at the problems of existing photovoltaic component defect detection methods and provides a photovoltaic panel defect detection method and system based on an infrared camera mounted on a drone. First, the EL image of the photovoltaic panel is collected by the infrared camera mounted on the drone, and an image processing algorithm model is built to identify abnormal areas on the photovoltaic panel and classify the faults. Then, artificial intelligence technology is used to accurately locate the abnormal areas on the photovoltaic panel. Finally, an edge-cloud collaborative strategy is adopted to achieve lightweight deployment, providing an unmanned non-destructive detection and identification method that can quickly and accurately detect photovoltaic panel defects.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is: a photovoltaic panel defect detection method based on an unmanned aerial vehicle equipped with an infrared camera, comprising the following steps:

[0009] S1, data collection: collecting data information of the photovoltaic panel through the infrared camera carried by the drone, wherein the data information at least includes high-definition images and EL image data of the surface of the photovoltaic panel;

[0010] S2, model building: building an image processing algorithm model to identify abnormal areas on photovoltaic panels and classify faults;

[0011] S3, precise positioning: for the abnormal area identified in step S2, the abnormal area on the photovoltaic panel is precisely positioned using artificial intelligence technology;

[0012] S4, edge-cloud collaboration strategy: The edge-cloud collaboration strategy is adopted to achieve lightweight deployment, thereby realizing photovoltaic panel defect detection.

[0013] As an improvement of the present invention, the image processing algorithm model in step S2 is a yolov8 model, and the EL data is identified and classified in combination with the mobile network attention mechanism and the channel attention mechanism, wherein the data set is divided in a random manner, and the test set is not less than 10%. Specifically, the yolov8n model is adopted, and the depth, width and maximum number of channels of the model are 0.33, 0.25, and 1024. The model has 21 layers, and the 0th, 1st, 3rd, 5th, 7th, 16th, and 19th layers are convolution layers, and the convolution kernels are 64, 128, 256, 512, and 1024 respectively. The 2nd, 4th, 6th, 8th, 12th, 15th, 18th, and 21st layers are C2f modules for fusing low-level feature maps with high-level feature maps. The 9th layer is a pooling layer with a pooling kernel size of 5; the 10th and 13th layers are upsampling layers for expanding the length and width of the feature map to twice the original length and width; the 11th and 14th layers are connection layers; the The 20th layer is the output layer, which is responsible for outputting the detection results; the input of the 0th layer is 640*640*3, where 64 is the number of channels and 3 is the convolution kernel size, and the output is 320*320*64; the output of the 1st layer is 160*160*128; the output of the 3rd layer is 80*80*256; the output of the 5th layer is 40*40*512; the output of the 7th layer is 20*20*1024; the output of the 10th layer is 40*40*1024; the output of the 1st layer is The output is 80*80*512; the output of the 14th layer is 80*80*768; the output of the 15th layer is 80*80*256; the output of the 16th layer is 40*40*256; the output of the 17th layer is 40*40*768; the output of the 18th layer is 40*40*512; the output of the 19th layer is 20*20*512; the output of the 20th layer is 20*20*1536; the output of the 21st layer is 20*20*1024.

[0014] As another improvement of the present invention, in step S3, calibration is performed based on OPENCV to achieve accurate positioning of abnormal areas of photovoltaic panels. The calibration process at least includes determining a calibration board, corner point detection, estimating parameters and coordinate conversion, wherein the calibration board is composed of 6X9 small squares of the same size, and each square has easy-to-detect marking points at the four corners. These marking points are key reference points for determining the viewing angle and position of the camera, wherein the size and shape of the squares are consistent, and the color of the squares is black and white, ensuring that the colors of adjacent squares are different from each other, so as to ensure that these marking points can be clearly photographed under various lighting and angles to complete corner point detection. The specific steps of calibration are as follows: first, the calibration board is photographed at different angles with a camera, and then the feature points in the image are detected to obtain the pixel coordinate values ​​of the calibration corner points, and the physical coordinate values ​​of the calibration board are calculated based on the known size of the calibration board and the origin of the world coordinate system, and then the internal parameter matrix A and the external parameter matrix A are solved according to the physical coordinate values ​​and the pixel coordinate values. In addition, the estimated parameters and coordinate transformation are calculated in the northeast celestial coordinate system. The world coordinates are (Xw, Yw, Zw), the camera coordinate system is (Xc, Yc, Zc), the image coordinates taken by the camera are (x, y), and the pixel coordinates in the image are (u, v). The rigid body transformation is used to transform the world coordinates into the camera coordinates. The rigid body transformation parameter matrix is ​​the external parameter matrix, and its matrix expression is: Where R is the rotation matrix and T is the translation matrix. The camera coordinates are converted to image coordinates using perspective projection. The conversion matrix is: Where f is the focal length; the image coordinates are converted to pixel coordinates using a secondary transformation, and the transformation matrix is Among them, α, β, γ, u0, v0 are all internal parameters, and the transformation matrix is ​​the internal parameter matrix; the overall coordinate transformation expression is:

[0015]

[0016] As another improvement of the present invention, the edge-cloud collaboration strategy of step S4 includes a drone cluster, an edge computing node, and a cloud computing platform;

[0017] The drone cluster: uses intelligent path planning and formation control technology to cover the EL map of the photovoltaic module site, and transmits the EL map to the edge computing node in real time; wherein, the path planning is based on the geographical location, distribution range and scheduled inspection tasks of the photovoltaic modules. The drone cluster performs intelligent path planning while considering the flight speed, endurance, obstacle avoidance ability of the drone, and the arrangement and spacing of the photovoltaic modules to ensure that the drone can efficiently and comprehensively cover all photovoltaic modules, and the generated path should include parameters such as flight altitude, direction, speed, etc. to ensure the shooting quality and reduce the flight time; the formation control technology dynamically adjusts the position and speed of the drone according to the flight environment and task requirements to ensure that the drone cluster maintains the formation and spacing during the flight to avoid mutual interference and collision, and the drones transmit data and synchronize the status through wireless communication to maintain overall consistency; coverage shooting refers to the drone flying on the planned path and using the infrared camera on board to shoot the photovoltaic modules to obtain electroluminescent (EL) images. When shooting, the drone needs to maintain a stable flight attitude and altitude to ensure that the EL image taken is clear and distortion-free;

[0018] The edge computing node: after receiving the raw data transmitted by the drone at the edge end, pre-processes the data, and uploads the EL image after passing through the edge computing point to the cloud computing platform; the data pre-processing includes checking and verifying the integrity, accuracy and synchronization of the data, as well as denoising, enhancing contrast, correcting distortion, etc. of the original EL image, and then accurately and stably uploading the processed data to the cloud platform;

[0019] The cloud computing platform: collects data after screening by edge nodes, performs in-depth analysis through machine learning models, realizes defect detection, and formulates or optimizes inspection strategies based on comprehensive data analysis in the cloud; the specific processing process of the cloud computing platform includes data collection, in-depth analysis, defect detection and comprehensive data analysis; data collection refers to the cloud computing platform collecting pre-processed EL image data uploaded by edge computing nodes, and storing and managing them to ensure the reliability and security of data; in-depth analysis refers to the use of yolov8's model to accurately extract image features and identify and classify defects such as cracks, black cores, and fragments of photovoltaic modules; defect detection refers to further analysis of the results of the yolov8 model to obtain the specific type of defects and the location, size and other information of the defects for subsequent processing and maintenance; comprehensive data analysis is to statistically analyze the inspection results of photovoltaic modules to obtain the number of occurrences, types and distribution patterns of photovoltaic module defects in order to optimize the optimization strategy.

[0020] As another improvement of the present invention, the edge computing point deploys a lightweight AI algorithm model to analyze the preprocessed data and identify potential defect areas, and dynamically adjust the allocation of computing resources.

[0021] As a further improvement of the present invention, the inspection strategy in the cloud computing platform in step S4 at least includes the flight path of the drone cluster and the EL graph optimization detection algorithm.

[0022] In order to achieve the above purpose, the present invention also adopts a technical solution: a photovoltaic panel defect detection system based on a drone equipped with an infrared camera, comprising at least an acquisition module, a construction module and a deployment module.

[0023] The acquisition module is used to acquire the EL graph of the distributed photovoltaic components and perform screening and digital preprocessing;

[0024] The construction module is used to build a YOLOV8 model, optimize the hyperparameters of the model, and train the model using the data in the training set to obtain a trained model;

[0025] The deployment module adopts an edge-cloud collaboration strategy to achieve efficient collaborative processing and lightweight deployment of drone clusters, edge computing nodes and cloud computing platforms.

[0026] Compared with the prior art, the present invention has the following beneficial effects: the present invention discloses a photovoltaic panel defect detection method and system based on an infrared camera carried by a drone, optimizes the hyperparameters of the YOLOV8 model, improves the performance of photovoltaic component abnormality detection using image detection, and realizes remote identification of hot spot effects, cracks and other performance degradation signs in photovoltaic panel arrays; optimizes the performance of locating abnormal positions of photovoltaic components, and realizes accurate defect positioning without the need for close-range manual inspection; optimizes the lightweight deployment scheme for photovoltaic component abnormality detection, and realizes remote management, safety maintenance and performance optimization of large-scale photovoltaic power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of the photovoltaic panel defect detection method based on the unmanned aerial vehicle equipped with an infrared camera according to the present invention;

[0028] Figure 2 This is a result diagram of photovoltaic panel defect detection in Example 1 of the present invention;

[0029] Figure 3 This is the best result diagram of photovoltaic panel defect detection in Example 1 of the present invention;

[0030] Figure 4 The positioning loss, confidence loss, classification loss, precision rate, and recall rate curves of the YOLOV8 model in Example 1 of the present invention;

[0031] Figure 5 Schematic diagram of positioning the calibration plate in step S3 of embodiment 1 of the present invention;

[0032] Figure 6 Schematic diagram of the navigation coordinate system (northeast celestial coordinate system) in step S3 of embodiment 1 of the present invention;

[0033] Figure 7 Schematic diagram of the drone coordinate system in step S3 of embodiment 1 of the present invention;

[0034] Figure 8 The coordinate information diagram of the upper left positioning point and the lower right positioning point of the drone viewing angle with the threshold value set to 0-20 in step S3 of embodiment 1 of the present invention;

[0035] Fig. 9 This is a coordinate information diagram of the conversion between the upper left positioning point of the drone perspective and the center point of the world perspective output in step S3 of embodiment 1 of the present invention. DETAILED DESCRIPTION

[0036] In order to make the above-mentioned purposes, features and advantages of the present invention more understandable, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0037] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0038] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0039] Example 1

[0040] A photovoltaic panel defect detection method based on an infrared camera mounted on a drone, such as Figure 1 As shown, at least the following steps are included:

[0041] S1, data collection: collecting data information of the photovoltaic panel through the infrared camera carried by the drone, wherein the data information at least includes high-definition images and EL image data of the surface of the photovoltaic panel;

[0042] Drone swarms are used for data collection. They are equipped with high-resolution cameras, infrared imagers and other sensors. They take comprehensive photos of photovoltaic panels according to the planned path, collect high-definition images and EL image data of the photovoltaic panel surface for subsequent defect analysis. The use of drone swarms has the advantage of flexible operation. Through intelligent path planning and formation control technology, they can autonomously navigate in complex environments, quickly cover large areas of photovoltaic panel arrays, and reduce detection blind spots.

[0043] S2, model building: building an image processing algorithm model to identify abnormal areas on photovoltaic panels and classify faults;

[0044] The image processing algorithm model refers to the use of the yolov8 model combined with the mobile network attention mechanism and the channel attention mechanism to identify and classify EL images, thereby realizing automatic detection and classification of faults. The data set is divided randomly, and the test set is not less than 10%. The yolov8n model is used. The depth, width and maximum number of channels of the model are 0.33, 0.25 and 1024 respectively. The model has 21 layers. The 0th, 1st, 3rd, 5th, 7th, 16th and 19th layers are convolution layers. The convolution kernels are 64, 128, 256, 512 and 1024 respectively. The 2nd, 4th, 6th, 8th, 12th, 15th, 18th and 21st layers are C2f modules, which are used to fuse low-level feature maps with high-level feature maps. The 9th layer is the pooling layer with a pooling kernel size of 5. The 10th and 13th layers are upsampling layers, which are used to expand the length and width of the feature map to twice the original size. The 11th and 14th layers are connection layers. The 20th layer is the output layer, which is responsible for outputting the detection results. The input of the 0th layer is 640*640*3. 64 is the number of channels, 3 is the convolution kernel size, the output is 320*320*64; the output of the first layer is 160*160*128; the output of the third layer is 80*80*256; the output of the fifth layer is 40*40*512; the output of the seventh layer is 20*20*1024; the output of the tenth layer is 40*40*1024; the output of the first layer is 80*80*512; the output of the 14th layer is The output is 80*80*768; the output of the 15th layer is 80*80*256; the output of the 16th layer is 40*40*256; the output of the 17th layer is 40*40*768; the output of the 18th layer is 40*40*512; the output of the 19th layer is 20*20*512; the output of the 20th layer is 20*20*1536; the output of the 21st layer is 20*20*1024.

[0045] The results of the model recognition and classification are as follows Figure 2 As shown in Table 1 below;

[0046] Table 1. YOLO model confusion matrix

[0047]

[0048] The types of photovoltaic panels include cracks, black cores, and fragments. Linear cracks, star cracks, fi nger in terruptions, and scratches are all classified as cracks; thick lines are also included in black cores; and corners are also included in fragments. Figure 3 This is the best result diagram of photovoltaic panel defects. Figure 2 and Figure 3 The box in the result graph indicates the location where the defect is detected, and the upper left corner of the box indicates the type of defect and the probability of that type. This method can quickly and accurately identify the fault type of the photovoltaic panel. The positioning loss, confidence loss, classification loss, precision rate, and recall rate curves of the YOLOV8 model are shown in the figure. Figure 4 As shown in the figure; the positioning loss function is a mathematical function used in the target detection model to measure the difference between the predicted bounding box and the true bounding box. It guides the model training process by calculating the distance or overlap between the two, so as to more accurately locate the target object, reduce prediction errors, and improve detection accuracy; the confidence loss is used to measure the difference between the model's confidence in whether the target object exists in the predicted box and the actual label, so that the model can more accurately judge whether the target object exists in the predicted box, thereby improving the detection accuracy; the classification loss is used to measure the difference between the model's prediction of the target object category and the actual label. By optimizing the classification loss, the model can more accurately classify the target object, thereby improving the classification accuracy. Precision refers to the proportion of truly positive samples among all instances predicted by the model as positive samples. The calculation formula is: Precision = TP / (TP+FP), where TP represents true positive examples (True Positive) and FP represents false positive examples (False Positive). Precision reflects the model's ability to identify positive samples. Recall refers to the proportion of all instances that are actually positive samples that are correctly predicted by the model as positive samples. The calculation formula is: Recal l = TP / (TP+FN), where FN represents false negative examples (False Negative). Recall reflects the model's coverage of positive samples.

[0049] S3, precise positioning: for the abnormal area identified in step S2, the abnormal area on the photovoltaic panel is precisely positioned using artificial intelligence technology;

[0050] Calibration based on OPENCV is used to achieve accurate positioning of abnormal areas of photovoltaic panels. The calibration process at least includes determining the calibration board, corner point detection, parameter estimation and coordinate conversion. The calibration board is composed of 6X9 small squares of the same size. There are easy-to-detect marking points on the four corners of each square. These marking points are key reference points used to determine the camera's viewing angle and position. The size and shape of the squares are consistent, and the colors of the squares are black and white. The colors of adjacent squares are different to ensure that these marking points can be clearly photographed under various lighting and angles to complete corner point detection. The specific steps of calibration are as follows: First, use a camera to shoot the calibration board at different angles, then detect the feature points in the image, obtain the pixel coordinate values ​​of the calibration corner points, and calculate the physical coordinate values ​​of the calibration board based on the known calibration board size and the origin of the world coordinate system, and then solve the intrinsic parameter matrix A and the extrinsic parameter matrix based on the physical coordinate values ​​and pixel coordinate values. In addition, the estimated parameters and coordinate transformation are calculated in the northeast celestial coordinate system. The world coordinates are (Xw, Yw, Zw), the camera coordinate system is (Xc, Yc, Zc), the image coordinates taken by the camera are (x, y), and the pixel coordinates in the image are (u, v). The rigid body transformation is used to transform the world coordinates into the camera coordinates. The rigid body transformation parameter matrix is ​​the external parameter matrix, and its matrix expression is: Where R is the rotation matrix and T is the translation matrix. The camera coordinates are converted to image coordinates using perspective projection. The conversion matrix is: Where f is the focal length; the image coordinates are converted to pixel coordinates using a secondary transformation, and the transformation matrix is Among them, α, β, γ, u0, v0 are all internal parameters, and the transformation matrix is ​​the internal parameter matrix; the overall coordinate transformation expression is:

[0051] The accurate positioning of abnormal areas of photovoltaic panels is based on OPENCV calibration. The calibration board is as follows Figure 5 As shown; the EL graph includes a training set and a test set, where the data set is divided randomly and the test set is not less than 10%; the spatial coordinate system calibrated based on OPENCV includes the world coordinate system, the camera coordinate system, the EL graph physical coordinate system and the pixel coordinate system; the navigation coordinate system (northeast sky coordinate system) and the drone coordinate system are as shown Figure 6 and Figure 7 The calibration process based on OPENCV includes determining the calibration plate, corner point detection, estimating internal parameters, and distortion correction. The coordinate information of the upper left and lower right positioning points of the drone's perspective with a threshold set to 0-20 is shown in the figure below. Figure 8 The corresponding output of the conversion coordinate information of the upper left positioning point of the drone's view and the center point of the world view is shown in Fig. 9 shown.

[0052] S4, edge-cloud collaboration strategy: The edge-cloud collaboration strategy is adopted to achieve lightweight deployment, thereby realizing photovoltaic panel defect detection.

[0053] The overall edge-cloud collaboration strategy is divided into three parts: drone clusters, edge computing nodes, and cloud computing platforms. The drone cluster uses intelligent path planning and formation control technology to capture EL images of photovoltaic modules on site, and transmits the EL images to edge computing nodes in real time to ensure the freshness of the data and the timeliness of processing. The edge computing nodes need to pre-process the data. After receiving the raw data transmitted by the drone at the edge, they will perform preliminary processing such as data cleaning, format conversion, and image noise reduction to prepare for further analysis. This method has the advantages of fast analysis, resource optimization, and timing feedback. The lightweight AI algorithm model deployed at the edge computing point can quickly analyze the pre-processed data and identify potential defective areas. In addition, the deployment of the AI ​​algorithm model at the edge computing point can dynamically adjust the allocation of computing resources while monitoring the status of photovoltaic modules in real time, greatly reduce the pressure on the cloud, and optimize local storage and processing capabilities. The EL graph after the edge computing point will be uploaded to the cloud computing platform, where high-quality data screened by edge nodes will be collected and in-depth analysis will be performed using powerful computing power and complex machine learning models to further improve the accuracy and reliability of defect detection. At the same time, based on comprehensive data analysis on the cloud, inspection strategies will be formulated or further optimized, including the flight path of drone clusters, EL graph optimization detection algorithms, etc., to further improve the overall detection efficiency. At the same time, the cloud provides a unified monitoring interface, allowing remote viewing of detection progress, results, and system status, and long-term storage of detection data to form historical archives for trend analysis and model performance evaluation, ultimately facilitating management and decision-making.

[0054] Example 2

[0055] The present embodiment also provides a photovoltaic panel defect detection system based on an infrared camera mounted on a drone, including: an acquisition module, used to obtain the EL diagram of distributed photovoltaic components, and perform screening and data preprocessing; an analysis module, used to perform correlation analysis on the data, decide whether to retain it based on the analysis results, and organize the retained operating data into high-dimensional array samples to produce a data set; a construction module, used to build a YOLOV8 model, optimize the model's hyperparameters, and use the data in the training set to train the model to obtain a trained model; a deployment module, which adopts an edge-cloud collaboration strategy to achieve efficient collaborative processing and lightweight deployment of drone clusters, edge computing nodes, and cloud computing platforms.

[0056] In addition, this embodiment also provides a computer device, which is suitable for a photovoltaic panel defect detection method based on an unmanned aerial vehicle equipped with an infrared camera and its application, including: a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the photovoltaic panel defect detection method based on an unmanned aerial vehicle equipped with an infrared camera as described in Example 1.

[0057] Finally, this embodiment also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method in any optional implementation of the above embodiment is executed. Wherein, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read On ly Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0058] It can be seen from the above that, in response to the high-efficiency, low-cost detection and operation and maintenance needs raised by the rapid development of photovoltaic power generation systems, the present invention provides an unmanned non-destructive detection and identification method that can quickly and accurately detect defects in photovoltaic panels, optimizes the hyperparameters of the YOLOV8 model, improves the performance of abnormal detection of photovoltaic components using image detection, realizes remote identification of hot spot effects, cracks and other signs of performance degradation in photovoltaic panel arrays, and realizes precise positioning of defects without the need for close manual inspection, which facilitates remote management, safety maintenance and performance optimization of large-scale photovoltaic power stations.

[0059] It should be noted that the above content only illustrates the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.

Claims

1. A photovoltaic panel defect detection method based on an infrared camera mounted on a drone, characterized in that: The steps include: S1, data collection: collecting data information of the photovoltaic panel through the infrared camera carried by the drone, wherein the data information at least includes high-definition images and EL image data of the surface of the photovoltaic panel; S2, model building: building an image processing algorithm model to identify abnormal areas on photovoltaic panels and classify faults; S3, precise positioning: for the abnormal area identified in step S2, the abnormal area on the photovoltaic panel is precisely positioned using artificial intelligence technology; S4, edge-cloud collaboration strategy: The edge-cloud collaboration strategy is adopted to achieve lightweight deployment, thereby realizing photovoltaic panel defect detection.

2. The photovoltaic panel defect detection method based on an unmanned aerial vehicle equipped with an infrared camera as claimed in claim 1, characterized in that: The image processing algorithm model in step S2 is the yolov8 model, which combines the mobile network attention mechanism and the channel attention mechanism to identify and classify the EL image data collected in step S1, wherein the data set is divided in a random manner, and the test set is not less than 10%; the depth, width and maximum number of channels of the yolov8n model are 0.33, 0.25, and 1024, respectively, and the model has 21 layers, and the 0th, 1st, 3rd, 5th, 7th, 16th, and 19th layers are convolutional layers; the 2nd, 4th, 6th, 8th, 12th, 15th, 18th, and 21st layers are C2f modules for fusing low-level feature maps with high-level feature maps; the 9th layer is a pooling layer; the 10th and 13th layers are upsampling layers for expanding the length and width of the feature map to twice the original length and width; the 11th and 14th layers are connection layers; the 20th layer is the output layer, which is responsible for outputting the detection results.

3. The photovoltaic panel defect detection method based on an unmanned aerial vehicle equipped with an infrared camera as claimed in claim 1, characterized in that: In step S3, calibration is performed based on OPENCV to accurately locate the abnormal area of ​​the photovoltaic panel. The calibration process at least includes determining the calibration plate, corner point detection, parameter estimation and coordinate conversion; photographing the calibration plate at different angles with a camera, detecting feature points in the image, obtaining pixel coordinate values ​​of the calibration corner points, and calculating the physical coordinate values ​​of the calibration plate based on the known size of the calibration plate and the origin of the world coordinate system, as well as the intrinsic parameter matrix A and the extrinsic parameter matrix 4. The photovoltaic panel defect detection method based on an unmanned aerial vehicle equipped with an infrared camera as claimed in claim 1, characterized in that: The edge-cloud collaboration strategy of step S4 includes a drone cluster, an edge computing node, and a cloud computing platform; The drone cluster: uses intelligent path planning and formation control technology to capture EL images of the photovoltaic module site, and transmits the EL images to the edge computing node in real time; the intelligent path planning is based on the geographical location, distribution range, and scheduled inspection tasks of the photovoltaic modules, the flight speed, endurance, obstacle avoidance ability of the drone, and the arrangement and spacing of the photovoltaic modules; the formation control technology dynamically adjusts the position and speed of the drone based on the flight environment and mission requirements; The edge computing node: performs data preprocessing on the raw data transmitted by the drone at the edge end, and uploads the EL image after passing through the edge computing point to the cloud computing platform; the data preprocessing at least includes checking and verifying the integrity, accuracy and synchronization of the data, denoising the original EL image, enhancing the contrast and correcting the distortion; The cloud computing platform: The cloud computing platform at least includes the steps of data collection, in-depth analysis, defect detection and comprehensive data analysis; Collect data filtered by edge nodes, conduct in-depth analysis through machine learning models, realize defect detection, and formulate or optimize inspection strategies based on comprehensive data analysis in the cloud.

5. The photovoltaic panel defect detection method based on an unmanned aerial vehicle equipped with an infrared camera as claimed in claim 2, characterized in that: The image feature size of the yolov8n model changes as follows: the input of the 0th layer is 640*640*3, where 64 is the number of channels, 3 is the convolution kernel size, and the output is 320*320*64; the output of the 1st layer is 160*160*128; the output of the 3rd layer is 80*80*256; the 5th layer is 40*40*512; the output of the 7th layer is 20*20*1024; the output of the 10th layer is 40*40*1024; The output of the first layer is 80*80*512; the output of the 14th layer is 80*80*768; the output of the 15th layer is 80*80*256; the output of the 16th layer is 40*40*256; the output of the 17th layer is 40*40*768; the output of the 18th layer is 40*40*512; the output of the 19th layer is 20*20*512; the output of the 20th layer is 20*20*1536; the output of the 21st layer is 20*20*1024.

6. The photovoltaic panel defect detection method based on an unmanned aerial vehicle equipped with an infrared camera as claimed in claim 3, characterized in that: The method for estimating parameters and coordinate conversion is specifically as follows: in the northeast celestial coordinate system, the world coordinates are (Xw, Yw, Zw), the camera coordinate system is (Xc, Yc, Zc), the image coordinates taken by the camera are (x, y), and the pixel coordinates in the image are (u, v); the rigid body transformation is used to transform the world coordinates into the camera coordinates, and the rigid body transformation parameter matrix is ​​the external parameter matrix, and its matrix expression is: Where R is the rotation matrix and T is the translation matrix. The camera coordinates are converted to image coordinates using perspective projection. The conversion matrix is: Where f is the focal length; Use the secondary transformation to convert the image coordinates into pixel coordinates. The transformation matrix is Among them, α, β, γ, u0, v0 are all internal parameters, and the transformation matrix is ​​the internal parameter matrix; the overall coordinate transformation expression is:

7. The photovoltaic panel defect detection method based on an unmanned aerial vehicle equipped with an infrared camera as claimed in claim 4, characterized in that: The edge computing point deploys a lightweight AI algorithm model to analyze the preprocessed data and identify potential EL defect areas, and dynamically adjust the allocation of computing resources.

8. The photovoltaic panel defect detection method based on an unmanned aerial vehicle equipped with an infrared camera as claimed in claim 4, characterized in that: The inspection strategy in the cloud computing platform in step S4 at least includes the flight path of the drone cluster and the EL graph optimization detection algorithm.

9. Photovoltaic panel defect detection system based on drone equipped with infrared camera, characterized by: At least including acquisition module, construction module and deployment module, The acquisition module is used to acquire the EL graph of the distributed photovoltaic components and perform screening and digital preprocessing; The construction module is used to build a YOLOV8 model, optimize the hyperparameters of the model, and train the model using the data in the training set to obtain a trained model; The deployment module adopts an edge-cloud collaboration strategy to achieve efficient collaborative processing and lightweight deployment of drone clusters, edge computing nodes and cloud computing platforms.

Images