Photovoltaic panel surface defect detection method and system based on physical property analysis

Abstract

The invention belongs to the technical field of photovoltaic panel surface defect detection, and discloses a photovoltaic panel surface defect detection method and system based on physical property analysis. The method comprises the following steps: firstly, acquiring a surface temperature distribution image, surface deformation data, ultrasonic echo data, a spectral image, spectral characteristic data and eddy current signal characteristic data of the photovoltaic panel by respectively utilizing an infrared thermal imager, a laser speckle interferometer, an ultrasonic flaw detector, a visible light multi-band imager and eddy current detection equipment; various abnormal regions such as temperature, deformation, ultrasonic echo, spectral characteristics and eddy current signals are determined; and then determining defect positions by integrating various abnormal regions, and determining the types and sizes of the surface defects of the photovoltaic panel by combining various data corresponding to the defect positions. According to the method, accurate positioning and identification of the surface defects of the photovoltaic panel are realized through a multi-physical property detection means, and the detection accuracy and reliability are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic panel surface defect detection, and in particular, to a method and system for detecting photovoltaic panel surface defects based on physical property analysis. Background Art

[0002] With the continuous growth of the global demand for clean energy, photovoltaic power generation, as a green and environmentally friendly way of obtaining energy, has been widely used and developed rapidly. As the core component of a photovoltaic power generation system, the performance and reliability of a photovoltaic panel directly affect the power generation efficiency and system life. However, during the production, transportation, and use of photovoltaic panels, various defects are likely to appear on the surface. Therefore, effective detection of photovoltaic panel surface defects is crucial.

[0003] Currently, single detection technologies are mainly used for photovoltaic panel surface defect detection. For example, some enterprises use visual inspection methods, relying on manual direct observation of the photovoltaic panel surface to identify obvious cracks, stains, and other defects; some also use simple optical imaging detection, taking images of the photovoltaic panel surface with an ordinary camera and using image processing algorithms to identify defects. In addition, some enterprises use single physical detection methods.

[0004] However, these existing single detection technologies have many problems. Visual inspection methods are greatly affected by human factors, with low detection efficiency and prone to missed detections; simple optical imaging detection is difficult to identify tiny defects and internal defects; single physical detection methods can only detect specific types of defects and cannot comprehensively cover all types of surface defects that may occur on photovoltaic panels, with insufficient detection accuracy and integrity, and it is difficult to meet the requirements of the photovoltaic industry for high-quality detection. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method and system for detecting photovoltaic panel surface defects based on physical property analysis to solve the above technical problems.

[0006] To achieve the above object, in a first aspect, a method for detecting photovoltaic panel surface defects based on physical property analysis is provided, which includes the following steps: Use an infrared thermal imager to perform a full-field scan on the photovoltaic panel to obtain a surface temperature distribution image of the photovoltaic panel, and determine a temperature anomaly area based on the surface temperature distribution image; Use a laser speckle interferometer to scan the surface of the photovoltaic panel to obtain surface deformation data of the photovoltaic panel, and determine a deformation anomaly area based on the surface deformation data; Use an ultrasonic flaw detector to perform ultrasonic scanning on the surface of the photovoltaic panel to obtain ultrasonic echo data of the photovoltaic panel surface, and determine an ultrasonic echo anomaly area based on the ultrasonic echo data; Use a visible light multi - band imager to obtain the spectral image of the photovoltaic panel surface, extract spectral feature data from the spectral image based on the spectral unmixing algorithm, and determine the spectral feature abnormal area according to the spectral feature data; Use an eddy current detection device to detect the surface area of the photovoltaic panel, obtain eddy current signal feature data, and determine the eddy current signal abnormal area based on the eddy current signal feature data; Determine the defect position according to the temperature abnormal area, the deformation abnormal area, the ultrasonic echo abnormal area, the spectral feature abnormal area and the eddy current signal abnormal area; Determine the photovoltaic panel surface defect recognition result including the type and size of the surface defect according to the surface temperature distribution image, the surface deformation data, the ultrasonic echo data, the spectral feature data and the eddy current signal feature data corresponding to the defect position.

[0007] In a second aspect, a photovoltaic panel surface defect detection system based on physical property analysis is provided. The system is used to execute the method described in the first aspect. The system includes: An infrared thermal imaging detection module, which is used to perform a full - area scan of the photovoltaic panel by using an infrared thermal imager, obtain the surface temperature distribution image of the photovoltaic panel, and determine the temperature abnormal area based on the surface temperature distribution image; A laser speckle deformation detection module, which is used to scan the surface of the photovoltaic panel by using a laser speckle interferometer, obtain the surface deformation data of the photovoltaic panel, and determine the deformation abnormal area based on the surface deformation data; An ultrasonic flaw detection module, which is used to perform an ultrasonic scan of the surface of the photovoltaic panel by using an ultrasonic flaw detector, obtain the ultrasonic echo data of the surface of the photovoltaic panel, and determine the ultrasonic echo abnormal area based on the ultrasonic echo data; A spectral feature analysis module, which is used to obtain the spectral image of the photovoltaic panel surface by using a visible light multi - band imager, extract spectral feature data from the spectral image based on the spectral unmixing algorithm, and determine the spectral feature abnormal area according to the spectral feature data; An eddy current signal detection module, which is used to detect the surface area of the photovoltaic panel by using an eddy current detection device, obtain eddy current signal feature data, and determine the eddy current signal abnormal area based on the eddy current signal feature data; A defect position determination module, which is used to determine the defect position according to the temperature abnormal area, the deformation abnormal area, the ultrasonic echo abnormal area, the spectral feature abnormal area and the eddy current signal abnormal area; A defect recognition analysis module, which is used to determine the photovoltaic panel surface defect recognition result including the type and size of the surface defect according to the surface temperature distribution image, the surface deformation data, the ultrasonic echo data, the spectral feature data and the eddy current signal feature data corresponding to the defect position.

[0008] The above technical solution has the following beneficial technical effects: Through the collaborative operation of an infrared thermal imager, a laser speckle interferometer, an ultrasonic flaw detector, a visible light multi-band imager, and an eddy current detection device, the present invention can comprehensively obtain multi-source data information such as the surface temperature, deformation, ultrasonic echo, spectrum, and eddy current signal of the photovoltaic panel. Compared with the drawback that a single detection technology can only capture specific types of defects, the organic combination of such multi-physical property detection means can cover all types of defects that may appear on the surface of the photovoltaic panel in all directions, avoiding missed detections and misdetections caused by a single detection dimension. On the basis of accurately positioning the defect location, through the comprehensive analysis of multiple data, the type and size of the surface defect can be accurately determined, providing a comprehensive and reliable basis for the quality assessment of the photovoltaic panel. In addition, the mode of collaborative detection by multiple devices realizes the automation and high efficiency of the detection process, reduces manual intervention, reduces the influence of human factors on the detection results, and improves the detection efficiency and stability. Description of the Drawings

[0009] The drawings are used to better understand the present invention and do not constitute an improper limitation to the present invention. Among them: Figure 1 is the overall flowchart of the method for detecting surface defects of a photovoltaic panel based on physical property analysis according to an embodiment of the present invention; Figure 2 is the specific flowchart of step S10 of an embodiment of the present invention; Figure 3 is the specific flowchart of step S20 of an embodiment of the present invention; Figure 4 is the specific flowchart of step S30 of an embodiment of the present invention; Figure 5 is the specific flowchart of step S40 of an embodiment of the present invention; Figure 6 is the specific flowchart of step S50 of an embodiment of the present invention; Figure 7 is the specific flowchart of step S60 of an embodiment of the present invention; Figure 8 is the specific flowchart of step S70 of an embodiment of the present invention; Figure 9 is the block diagram of a system for detecting surface defects of a photovoltaic panel based on physical property analysis according to an embodiment of the present invention; Figure 10 is the structural schematic diagram of the computer system according to an embodiment of the present invention. Detailed Embodiments

[0010] The following describes exemplary embodiments of the present invention in conjunction with the accompanying drawings. Various details of the embodiments of the present invention are included to facilitate understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, descriptions of well-known functions and structures are omitted in the following description for clarity and conciseness.

[0011] As Figure 1 shown, this embodiment provides a method for detecting surface defects of a photovoltaic panel based on physical property analysis, which includes the following steps: S10: Use an infrared thermal imager to perform a full-field scan of the photovoltaic panel, obtain the surface temperature distribution image of the photovoltaic panel, and determine the temperature anomaly area based on the surface temperature distribution image; First, place the photovoltaic panel to be detected in a constant temperature and humidity environment chamber with a temperature of 25°C and a humidity of 40% and let it stand for 30 minutes to make the surface temperature of the photovoltaic panel reach the environmental stable state, avoiding interference from environmental factors on the temperature detection results. Select an infrared thermal imager for detecting the surface temperature distribution of the photovoltaic panel. Before detection, calibrate the infrared thermal imager by measuring a blackbody radiation source with a known temperature and adjusting the device parameters to ensure the accuracy of temperature measurement. During detection, fix the infrared thermal imager on an adjustable tripod, adjust the device with a level to keep it perpendicular to the surface of the photovoltaic panel, and set the distance between the two to 0.5 meters. Turn on the infrared thermal imager, set the resolution to 640×480 pixels, the temperature measurement range to -20°C to 150°C, the accuracy to ±1°C, and the scanning speed to 10 frames per second, and perform a full-field scan of the photovoltaic panel to obtain the surface temperature distribution image. After obtaining the image, use the image processing software installed on the computer to process the image using a region growing algorithm based on threshold segmentation. This algorithm uses the average value of the normal operating temperature as a reference, sets the temperature deviation threshold to ±5°C, marks the areas where the temperature exceeds this range as seed points, and then starts from the seed points and merges adjacent pixels according to the similarity criterion (the temperature difference is within a certain range) to finally form the temperature anomaly area.

[0012] S20: Use a laser speckle interferometer to scan the surface of the photovoltaic panel, obtain the surface deformation data of the photovoltaic panel, and determine the deformation anomaly area based on the surface deformation data; In this step, a laser speckle interferometer is selected. It features high resolution and high sensitivity and can detect minute deformations on the surface of the photovoltaic panel. Before detection, the optical path of the laser speckle interferometer is calibrated. By adjusting the positions of the laser emitter and the receiver, it is ensured that the laser beam accurately irradiates the surface of the photovoltaic panel and can effectively receive the reflected light. The photovoltaic panel is fixed on a vibration table, and a sinusoidal excitation with a frequency of 50 Hz and an amplitude of 0.01 mm is applied to it to stimulate the minute vibrations on the surface of the photovoltaic panel, facilitating the detection of surface deformations. The laser wavelength of the laser speckle interferometer is set to 532 nm, the power is 20 mW, and the scanning area is set to a 10 cm × 10 cm grid on the surface of the photovoltaic panel. During the scanning process, the phase-shifting technique is used to obtain the interference fringe image. Through the four-step phase-shifting algorithm, that is, successively changing the phase of the laser to obtain four interference images with different phases, and then calculating the surface deformation data through the four-step phase-shifting algorithm. After obtaining the deformation data through calculation, the amount of deformation is compared with the preset threshold of 0.1 μm, and the area where the amount of deformation exceeds this threshold is determined as the deformation abnormal area.

[0013] S30: Use an ultrasonic flaw detector to perform ultrasonic scanning on the surface of the photovoltaic panel to obtain ultrasonic echo data on the surface of the photovoltaic panel, and determine the ultrasonic echo abnormal area based on the ultrasonic echo data; In this step, an ultrasonic flaw detector is used, equipped with a 5 MHz dual-crystal probe. The dual-crystal probe can effectively reduce the interference in the near field area and improve the accuracy and resolution of detection. Before detection, the ultrasonic flaw detector is calibrated for sound velocity. A standard test block with the same material as the photovoltaic panel is selected, and by measuring the propagation time of ultrasonic waves in the test block, the correct sound velocity parameter is calculated and set. During detection, a water-based coupling agent is evenly applied to the surface of the photovoltaic panel to ensure that ultrasonic waves can smoothly penetrate into the interior of the photovoltaic panel. The probe of the ultrasonic flaw detector is linearly scanned along the surface of the photovoltaic panel at a speed of 0.5 m / s, and the scanning pitch is set to 2 mm to ensure full coverage of the detection area. At the same time, the ultrasonic emission voltage is set to 100 V and the gain is 40 dB to obtain a clear ultrasonic echo signal. For the collected echo signal, using the signal processing software in the computer, the wavelet transform algorithm is used for noise reduction processing to remove the noise interference in the signal. Then, the amplitude, time, and frequency characteristics of the echo signal are extracted, and the area where the amplitude attenuation of the echo signal exceeds 50% or abnormal reflection peaks appear is marked as the ultrasonic echo abnormal area.

[0014] S40: Use a visible light multi-band imager to obtain the spectral image of the surface of the photovoltaic panel, extract spectral feature data from the spectral image based on the spectral unmixing algorithm, and determine the spectral feature abnormal area according to the spectral feature data; The spectral range of the visible light multi-band imager covers 400 - 1000 nm, with a spectral resolution of 5 nm, capable of acquiring rich spectral information. The detection is carried out under a standard D65 light source, which simulates the spectral distribution of average sunlight and can provide stable and uniform lighting conditions. Before detection, dark current calibration and radiometric calibration are performed on the imager to eliminate the device's own noise and improve the accuracy of spectral data. Fix the imager on a movable detection bracket, adjust its position and angle so that it can completely capture the surface of the photovoltaic panel. Set the exposure time to 100 ms, perform area scanning on the surface of the photovoltaic panel to obtain spectral images. A linear mixing model is used for spectral unmixing. This model is based on the principle of linear combination of spectral data and decomposes each pixel point in the spectral image into a combination of spectral characteristics of different substances. Compare the spectral feature vectors obtained from unmixing with a preset defect spectral library, which stores the standard spectral characteristics of different types of defects. By calculating the similarity between spectral feature vectors, when the similarity exceeds 85%, the corresponding area is determined as the spectral feature abnormal area.

[0015] S50: Use an eddy current detection device to detect the surface area of the photovoltaic panel, obtain eddy current signal characteristic data, and determine the eddy current signal abnormal area based on the eddy current signal characteristic data; Specifically, the eddy current detector is equipped with an absolute probe with an outer diameter of 8 mm. The absolute probe can detect the overall characteristic changes of the object to be detected and is suitable for the preliminary screening of surface defects of photovoltaic panels. Keep the probe of the eddy current detector perpendicular to the surface of the photovoltaic panel at a distance of 1 mm to ensure that the eddy current can effectively penetrate the surface layer of the photovoltaic panel. Set the excitation frequency to 10 kHz and the detection speed to 10 mm / s, and perform a grid-like scan on the surface area of the photovoltaic panel with a scan spacing of 1 mm. During the scan, eddy current signals are collected in real time, and the impedance change characteristics of the eddy current signals are extracted. Compare the collected impedance change data with the impedance data of the normal area, calculate the impedance change rate, and when the impedance change rate exceeds 30% of the normal area, mark the corresponding area as the eddy current signal abnormal area.

[0016] S60: Determine the defect position based on the temperature abnormal area, the deformation abnormal area, the ultrasonic echo abnormal area, the spectral feature abnormal area, and the eddy current signal abnormal area; The abnormal areas obtained by the above five detection methods are projected onto a unified coordinate system through Geographic Information System (GIS) technology. The weighted voting method is used to determine the defect location. To make the voting result more in line with the actual detection situation, different weights are set according to the detection sensitivity and accuracy of different detection methods for various types of defects: the weight of the temperature abnormal area is 0.2 because temperature detection is more sensitive to defects such as hot spots; the weight of the deformation abnormal area is 0.25, and deformation detection can effectively detect surface deformation defects; the weight of the ultrasonic echo abnormal area is 0.25, and ultrasonic flaw detection has a good effect on detecting internal defects; the weight of the spectral feature abnormal area is 0.2, and spectral detection can identify defects such as material changes; the weight of the eddy current signal abnormal area is 0.1, and eddy current detection is mainly used for preliminary screening of changes in surface conductivity. When a certain area is marked as abnormal in at least three detection methods and the comprehensive weight score exceeds 0.6, this area is determined as the defect location. The specific calculation method is to convert the abnormal marks of this area in each detection method into scores (1 for abnormal mark and 0 for no abnormal mark), then multiply by the corresponding weights respectively, and finally sum to obtain the comprehensive weight score.

[0017] S70: Determine the photovoltaic panel surface defect recognition result including the type and size of surface defects according to the surface temperature distribution image, surface deformation data, ultrasonic echo data, spectral feature data, and eddy current signal feature data corresponding to the defect location.

[0018] For the determined defect location, collect its corresponding five types of physical property data, namely surface temperature distribution image, surface deformation data, ultrasonic echo data, spectral feature data, and eddy current signal feature data. Establish a defect type classification model based on Support Vector Machine (SVM). Before constructing the model, first normalize the five types of data and map the data uniformly to the interval [0, 1] to eliminate the influence of data dimensions. Use the normalized feature vectors as the input of the SVM model, and the output is the defect type, including cracks, delamination, bubbles, contamination, etc. The SVM model is trained with a large number of known defect sample data to adjust the model parameters to make it have accurate classification ability. Adopt a morphological processing algorithm to calculate the area and perimeter of the defect area. The specific process is as follows: first, binarize the image of the defect area to convert it into a black-and-white image, then use morphological operations such as erosion and dilation to remove noise and fill holes, and finally calculate the area by counting the number of white pixel points in the image and calculate the perimeter through the boundary tracking algorithm. According to the preset conversion coefficient, convert the pixel size in the image into the actual physical size to determine the actual size of the defect. Finally, output a photovoltaic panel surface defect recognition result report including the defect type and size.

[0019] This solution comprehensively utilizes an infrared thermal imager, a laser speckle interferometer, an ultrasonic flaw detector, a visible light multi-band imager, and an eddy current detection device to obtain data from physical property dimensions such as temperature, deformation, ultrasonic echo, spectral characteristics, and eddy current signals respectively. It can comprehensively cover various defects that may appear on the surface of photovoltaic panels, such as cracks, delamination, bubbles, and contamination, avoiding the limitations of single detection technologies that can only target specific defects and reducing the risk of missed detection. In terms of detection accuracy, each detection device is optimized and calibrated according to its characteristics, and is combined with corresponding data processing algorithms, such as the threshold segmentation region growing algorithm for infrared thermal imaging detection and the four-step phase-shift algorithm for laser speckle interferometry detection. It can accurately capture defect features, and at the same time determine the defect location through the weighted voting method, classify the defect types using a support vector machine, and calculate the defect size by combining morphological processing algorithms to achieve high-precision identification of the defect location, type, and size. In terms of detection efficiency and reliability, the automated detection process reduces manual intervention and human error. At the same time, the collaborative work of multiple devices and the integration of multiple algorithms improve the stability and reliability of detection.

[0020] As Figure 2 shown, in some embodiments, step S10 specifically includes: S11: Set the scanning range of the infrared thermal imager to cover the entire surface of the photovoltaic panel, perform a full-field scan of the photovoltaic panel, and obtain multiple frames of surface temperature distribution images; Place the photovoltaic panel to be detected in a darkroom environment or outdoor environment with a temperature of 25°C ± 0.5°C and a humidity of 40% ± 5% and let it stand for 30 minutes to ensure uniform surface temperature. Select an infrared thermal imager (equipped with a macro lens, field of view 45°×34°), fix it 1.5 meters directly above the photovoltaic panel through an electric lifting platform, and adjust the pan-tilt head so that the optical axis is perpendicular to the panel surface. Perform two-point blackbody calibration (25°C / 50°C) before detection, set the parameters as resolution 640×480 pixels, frame rate 20Hz, emissivity 0.95, temperature measurement range 20°C - 80°C, and thermal sensitivity less than 0.03°C. Adopt a spiral scanning mode (starting point is the center of the panel, pitch 5cm), scanning speed 0.1m / s, continuously collect 60 frames of images, and ensure that the overlap rate of each frame is ≥30%. Immediately perform time synchronization and geotagging after image acquisition, and store the original data in an industrial control computer.

[0021] S12: Use the adaptive threshold filtering algorithm to perform noise reduction processing on the surface temperature distribution images to obtain preprocessed surface temperature distribution images; Import the collected image sequence into the MATLAB R2023a environment. First, perform time-domain filtering (5-frame median filtering) to remove random noise. The adaptive Wiener filtering algorithm is adopted, and the specific implementation is as follows: Divide the image into overlapping sub-blocks of 8×8 pixels; calculate the local mean and variance of each sub-block; calculate the frequency-domain transfer function according to the formula H(u,v)=1 / [1+σ² / σε²(u,v)], where σ² is the noise variance (estimated through the dark area of the image), and σε²(u,v) is the local variance; perform frequency-domain filtering on each sub-block and then reconstruct the image. Finally, output a preprocessed image with SNR≥38dB.

[0022] S13: Calculate the temperature deviation between the temperature values of each pixel point in the preprocessed surface temperature distribution image and the preset normal temperature range, and mark the pixel points with temperature deviations exceeding the dynamic threshold as abnormal points; First, construct a temperature reference model, perform Gaussian smoothing (σ = 1.5) on the preprocessed image, and calculate the global temperature mean μ0 = 30.2°C and standard deviation σ0 = 2.1°C. Set the normal temperature range as [μ0 - 2σ0, μ0 + 2σ0], that is, [26.0°C, 34.4°C]. The dynamic threshold calculation adopts an iterative optimization strategy as follows: (1) The initial threshold τ0 = μ0 + 1.5σ0 = 33.4°C; (2) Mark the pixels with temperatures greater than T0 as candidate abnormal points; (3) Calculate the temperature mean μ1 and standard deviation σ1 of the candidate points; (4) Update the threshold τ1 = μ1 + 0.8σ1; (5) Repeat steps 2 to 4 until τ converges (the change is less than 0.1°C). Finally, determine the threshold τ = 35.7°C, mark the pixel points with temperatures greater than τ as abnormal points, and generate a binary marking map (the value of abnormal points is 255, and the value of normal points is 0).

[0023] S14: Perform connectivity analysis on the pixel points marked as abnormal points, use the connected region algorithm to group adjacent abnormal points into the same abnormal sub-region, and generate a preliminary region map of the temperature abnormal region; within each abnormal sub-region of the preliminary region map, adopt the region expansion algorithm to further connect adjacent but not yet connected abnormal points according to the preset spatial distance threshold to form a temperature abnormal region with continuous boundaries.

[0024] The generation of connected regions adopts a parallel queue filling algorithm, and the specific steps are as follows: Divide the marking map evenly into 16×16 processing blocks, and each processing block is independently processed in parallel; perform 4-neighborhood connectivity analysis on the abnormal points within each processing block to identify the local connected regions within the block; merge the cross-block regions through the boundary matching algorithm, and use a region ID mapping table to record the region merging relationship to ensure that the same connected region in adjacent processing blocks is assigned the same ID.

[0025] The implementation process of the region expansion algorithm includes the following steps: Calculate the geometric center and circumscribed rectangle of each connected region for subsequent spatial relationship analysis; Screen adjacent region pairs with a distance less than 15 pixels and a temperature gradient greater than 0.5 °C / pixel, and construct Delaunay triangulation for the qualified region pairs to identify connection paths; Search for bridge points with a temperature greater than 34.5 °C in the triangular grid, and these points serve as key nodes for region connection; Use the region growing method to connect adjacent regions, with the nearest boundary point of the region as the seed point, and the growth condition is set to a temperature greater than 34 °C and meeting the 8-neighborhood connectivity requirement.

[0026] The finally generated abnormal region map contains complete boundary information, and calculates 12 geometric feature parameters of each region, specifically including area, perimeter, eccentricity, circularity, rectangularity, elongation, compactness, equivalent diameter, orientation angle, second moment, Euler number, and fractal dimension.

[0027] By setting the scanning range of the infrared thermal imager to full coverage and obtaining multiple frames of images, detection blind spots can be effectively avoided. At the same time, the multiple frames of data provide redundant information for subsequent analysis, improving the detection reliability; The adaptive threshold filtering algorithm can dynamically adjust the noise reduction parameters according to the local features of the image. Compared with the fixed threshold method, it can remove noise more accurately, retain the temperature anomaly details to the greatest extent, and improve the image quality; The method of dynamically thresholding to mark abnormal points fully considers the working environment of the photovoltaic panel and the measurement error of the equipment, and is more in line with the actual situation than the static threshold, effectively reducing the probability of false detection and missed detection; The combination of connectivity analysis and region expansion algorithm first preliminarily divides the abnormal sub-regions through the connected region algorithm, and then expands the regions according to the spatial distance threshold. It not only quickly locates the abnormal regions, but also can integrate adjacent small abnormal points into a complete defect region, accurately define the defect boundary, and finally achieve high-precision and high-efficiency detection and identification of the temperature abnormal regions on the surface of the photovoltaic panel.

[0028] As Figure 3 shown, step S20 specifically includes: S21: Control the laser speckle interferometer to scan the surface of the photovoltaic panel with pulsed laser at a preset frequency. During the scanning process, use the spatial light modulator in the laser speckle interferometer to modulate the laser speckle pattern in real time, so that the laser speckle pattern forms a dynamically changing modulated speckle field on the surface of the photovoltaic panel; Fix the photovoltaic panel to be detected horizontally on a vibration isolation platform (vibration suppression rate greater than 90%), and adjust the position of the laser speckle interferometer to make it perpendicular to the surface of the photovoltaic panel, with a distance of 0.8 meters. Turn on the laser emitter, set the pulse frequency to 10 kHz, the laser wavelength to 532 nm, and the power to 15 mW. Control the spatial light modulator to modulate the laser speckle pattern in real time. The specific modulation method is as follows: Generate a pseudo-random phase mask sequence through an FPGA (Field-Programmable Gate Array) controller. The size of each phase mask is 1024×768 pixels, and the phase modulation depth is 0 - 2π. Load the phase mask sequence onto the spatial light modulator at a frequency of 50 Hz, so that a dynamically changing modulated speckle field is formed on the surface of the photovoltaic panel. During the modulation process, the laser power stability is monitored in real time through a closed-loop control system to ensure that the power fluctuation is less than ±2%.

[0029] The FPGA controller is connected to the spatial light modulator through a high-speed data interface. Its function is to generate a pseudo-random phase mask sequence and load the phase mask onto the spatial light modulator at a frequency of 50 Hz to achieve real-time modulation of the laser speckle pattern. At the same time, it also receives the laser power monitoring data from the photodiode, and dynamically adjusts the laser drive current through the built-in PID control algorithm and closed-loop control system to ensure that the laser power fluctuation is less than ±2%. In addition, the FPGA controller also generates synchronization signals to control the pulsed laser and the high-speed camera respectively, ensuring the precise timing synchronization of laser emission, speckle field modulation, and image acquisition.

[0030] S22: Collect a sequence of modulated speckle images corresponding to the modulated speckle field on the surface of the photovoltaic panel by the camera at a frame rate synchronized with the pulsed laser; Use a high-speed camera to synchronously collect modulated speckle images with the pulsed laser. Achieve the synchronization of the camera and the laser through a trigger signal generator. Set the camera frame rate to 50 fps (matching the phase mask update frequency), the exposure time to 10 μs, and the aperture to F5.6. During the collection process, use an ambient light suppression device (narrowband filter with a central wavelength of 532 nm and a bandwidth of 10 nm) to reduce ambient light interference. Continuously collect 100 frames of modulated speckle images to form a complete image sequence. After each frame of image is collected, it is immediately transmitted to the industrial control computer through Gigabit Ethernet for caching to ensure the time continuity and spatial consistency of the image sequence.

[0031] S23: Process the sequence of modulated speckle images using a deep learning-based phase unwrapping algorithm to obtain the restored phase information, and extract the surface deformation data of the photovoltaic panel from the restored phase information; The acquired modulated speckle image sequence is input into the phase unwrapping algorithm module based on deep learning. This module adopts an improved U-Net network architecture and is specifically implemented as follows: The data preprocessing unit performs normalization on the input modulated speckle image sequence, converting the pixel values of each frame to the range [0, 1] through a linear mapping. This unit outputs a normalized image sequence with a uniformly constrained pixel value distribution, effectively eliminating data fluctuations caused by differences in lighting conditions and camera parameters between image batches.

[0032] The feature extraction unit extracts features from the normalized image sequence using an encoder network. The network consists of five convolutional blocks, each consisting of two 3×3 convolutional layers and a ReLU (Rectified Linear Unit) activation function. As the network depth increases, the feature map size is halved (from the initial 1280×800 to 80×50), while the number of channels is doubled (from 64 to 1024 channels). This feature extraction unit outputs a set of multi-scale feature maps, including intermediate feature maps from each encoder layer (for subsequent skip connections) and a final high-level abstract feature map. These feature maps capture key information in the speckle image, such as phase gradients and texture variations, providing a rich semantic representation for phase unwrapping. The intermediate feature maps are passed to the corresponding layers of the phase unwrapping unit, and the final feature map serves as the initial input to the decoder.

[0033] The phase unwrapping unit, used in a decoder network with skip connections, concatenates the intermediate feature maps of each encoder layer with the decoder feature maps of the corresponding layer, gradually recovering high-resolution phase information. The specific process involves upsampling the high-level feature maps output by the encoder (via transposed convolution or bilinear interpolation); channel-wise concatenating the upsampled feature maps with the intermediate feature maps of the corresponding encoder layer; fusing the concatenated features through a convolutional layer to generate a more accurate phase representation; and repeating the upsampling, concatenation, and feature fusion process until the original resolution phase map is restored. This unit ultimately outputs a continuous phase map with the same resolution as the input speckle image (1280×800 pixels) and a phase value range of [-π, π]. Each pixel value in the phase map corresponds to the phase information at that point on the photovoltaic panel surface and is directly passed to the deformation data extraction unit for physical quantity conversion.

[0034] During network training, a dataset containing 10,000 pairs of simulated speckle images and real phase maps was used for training. The weighted mean square error (W-MSE) was used as the loss function, and the weights were adaptively adjusted according to the phase gradient.

[0035] A deformation data extraction unit is used to calculate surface deformation data from the restored phase information according to the formula Δd = λ・φ / (4π・cosθ), where λ is the laser wavelength, φ is the phase change, and θ is the incident angle (in this embodiment, θ = 0°).

[0036] S24: Based on a preset deformation threshold, analyze the surface deformation data through a clustering analysis algorithm to determine the deformation abnormal area.

[0037] Import the extracted surface deformation data into the clustering analysis module, and the specific implementation is as follows: In the data preprocessing step, perform Gaussian filtering (σ = 1.5) on the deformation data to remove high-frequency noise. In the feature extraction step, calculate the statistical features of the deformation data, including mean, standard deviation, skewness, and kurtosis. In the clustering analysis step, use the DBSCAN algorithm (Density-Based Spatial Clustering of Applications with Noise) for anomaly detection, and set the parameters as neighborhood radius ε = 0.05μm and minimum number of points MinPts = 10. In the abnormal area determination step, mark the area that deviates from the main clustering center by more than 3 times the standard deviation in the clustering result as a potential abnormal area. In the threshold screening step, compare the maximum deformation amount of the potential abnormal area with the preset deformation threshold (0.1μm in this embodiment), and the area exceeding the threshold is determined as the deformation abnormal area. In the result verification step, perform morphological operations (dilation and erosion) on the determined deformation abnormal area to optimize the boundary, and finally generate a detection report including the position, area, and maximum deformation amount of the abnormal area.

[0038] As Figure 4 shown, step S30 specifically includes: S31: Control the ultrasonic flaw detector to perform ultrasonic scanning on the surface of the photovoltaic panel row by row, and synchronously collect the ultrasonic echo data of each scanning point; Fix the photovoltaic panel to be detected horizontally on the detection platform, adjust the position of the ultrasonic flaw detector to make it perpendicular to the surface of the photovoltaic panel, and the distance is 0.5 meters. Select a 5MHz dual-crystal probe (wafer size 10mm×10mm, focal length 50mm), and apply an appropriate amount of water-based coupling agent between the probe and the surface of the photovoltaic panel to ensure that ultrasonic waves can effectively penetrate into the photovoltaic panel. Set the parameters of the ultrasonic flaw detector as follows: emission voltage 100V, gain 40dB, sampling frequency 100MHz, sampling depth 100mm. Use an automatic scanning system to control the probe to scan along the surface of the photovoltaic panel at a speed of 0.5m / s row by row, the scanning pitch is 2mm, and the scanning width covers the entire surface of the photovoltaic panel. During the scanning process, synchronously collect the ultrasonic echo data of each scanning point. Each echo signal contains 1024 sampling points, and the data format is a time-domain waveform. The collected echo data is transmitted to the industrial control computer in real time through the USB interface and stored in the form of a binary file, and the file name contains the scanning position coordinate information.

[0039] S32: Input the collected ultrasonic echo data into a pre-trained convolutional neural network model to output the probability values of defects existing at each scanning point; After the collected ultrasonic echo data is input into a pre-trained convolutional neural network model, the model extracts the feature information in the data through the calculations and processes of each internal layer, and finally outputs the probability values of defects existing at each scanning point. The probability values range between [0, 1], and the closer the value is to 1, the higher the possibility that the scanning point has a defect. This model can adopt an improved ResNet-18 architecture, and its specific structure is as follows: Data preprocessing module. As the input front-end of the model, this module normalizes the original ultrasonic echo data and maps the amplitude of each sampling point to the interval [-1, 1]. This operation can eliminate the signal intensity differences caused by factors such as equipment fluctuations and material differences between different scanning points, standardize the data distribution, provide a unified and standardized data basis for subsequent feature extraction, and enhance the stability of model training and prediction.

[0040] Feature extraction module, which is composed of a 1D convolutional layer, a batch normalization layer, and a ReLU activation function connected in sequence. The 1D convolutional layer contains 64 convolutional kernels with a size of 7×1, and performs a sliding convolution operation on the normalized echo data with a stride of 2. Through the local calculation of the convolutional kernels and the data, the time-domain features of the echo signal are extracted to generate 64 feature maps. Subsequently, the batch normalization layer normalizes the feature maps, accelerates the model convergence, and reduces the risk of overfitting. The ReLU activation function introduces non-linearity into the model and enhances the model's ability to express complex features. The feature maps output by this module are passed to the residual block design module.

[0041] Residual block design module, which is composed of multiple cascaded residual blocks. Each residual block contains two 3×1 convolutional layers inside, and a batch normalization layer and a ReLU activation function are interspersed between the two convolutional layers. The residual block uses a residual connection structure to directly add the input to the output of the convolutional layer, enabling the network to directly learn the residual information by skipping some layers during the learning process. This design effectively solves the problem of gradient disappearance during the training of deep networks, allows the model to construct a deeper network structure, thereby extracting higher-level and more complex feature information, and improving the model's ability to capture subtle defect features. The residual blocks in the module process the feature maps in sequence, and finally output the feature maps after deep feature extraction to the global pooling and classification module.

[0042] Global pooling and classification module. First, through the global average pooling layer, the feature map output by the residual block is compressed into a one-dimensional vector, retaining key feature information while reducing the data dimension. Subsequently, a fully connected layer is connected. The fully connected layer calculates the one-dimensional vector according to the weight parameters learned from the training data, and finally outputs the probability values of defects existing at each scan point. This probability value reflects the prediction result of the model on whether there are defects at each scan point and is used for subsequent judgment of abnormal regions.

[0043] During model training, a dataset containing 10,000 samples is used as the training basis, including 5,000 defective samples (covering different types of defects such as cracks, delaminations, and bubbles) and 5,000 normal samples. During the training process, the Adam (Adaptive Moment Estimation) optimizer is used to iteratively update the model parameters at a learning rate of 0.001. Each time, 64 samples are selected to form a batch, and a total of 100 rounds of training are performed. The binary cross-entropy loss function is used to quantify the difference between the predicted probability value and the true label of the sample (1 for defective and 0 for normal). The model minimizes the loss function through the backpropagation algorithm, continuously adjusts the internal parameters, and maximizes the ability to distinguish between defective samples and normal samples, so as to achieve accurate identification and prediction of defect features in ultrasonic echo data.

[0044] S33: According to a preset probability threshold, determine the area where the scan points with probability values higher than the preset probability threshold as the ultrasonic echo abnormal area.

[0045] First, the probability of a defect at each scan point output by the convolutional neural network model is compared against a pre-set probability threshold (set to 0.7 in this example). Scanning points with a probability greater than 0.7 are marked as potential defect points, and their two-dimensional (X, Y) coordinates on the photovoltaic panel surface are recorded to form an initial set of potential defect points. Subsequently, these marked potential defect points are spatially clustered, and the density clustering algorithm (DBSCAN) is used to merge regions. Specifically, a neighborhood radius ε = 5 mm is set around each potential defect point, searching for other potential defect points within a 5 mm radius of the center point. Furthermore, a minimum number of points, MinPts = 3, is defined, meaning that if a point's neighborhood contains at least three potential defect points, these points are grouped together. The DBSCAN algorithm traverses all potential defect points, merging those with similar spatial distances that meet the density threshold to form multiple clusters. After clustering, each cluster is validated. The area of each region is calculated, and areas with an area less than 2mm² are eliminated (this threshold is set based on the minimum size of common defects in photovoltaic panels). Such areas are usually identified as misjudgments caused by noise interference. For the retained areas, their geometric features are further calculated, including parameters such as centroid coordinates, boundary perimeter, and aspect ratio. Finally, the clustered areas that have been screened and feature calculated are identified as ultrasonic echo abnormality areas. To facilitate subsequent analysis and recording, each abnormal area is assigned a unique number, and key information such as the area's coordinate range, area, and centroid coordinates is stored in the test result database. At the same time, in the visual interface, each abnormal area is marked with different colors to intuitively display the defect distribution on the photovoltaic panel surface.

[0046] The advantages of this technical solution are that, at the data acquisition level, it adopts line-by-line ultrasonic scanning and synchronously collects echo data at each scanning point to ensure full coverage of the photovoltaic panel surface. Combined with specific probe parameters and coupling agents, it ensures the effective transmission of ultrasonic waves and improves data quality. In the data analysis stage, it uses a pre-trained convolutional neural network model to process echo data, extracts time domain features through an improved ResNet-18 architecture, and uses residual connections to solve the gradient problem of deep networks. Compared with traditional analysis methods, it can capture subtle defect features and output the defect probability value of each scanning point. In terms of defect judgment, the scanning points are screened and divided into regions through the combination of preset probability thresholds and the DBSCAN clustering algorithm, which can not only effectively eliminate single misjudgment points, but also fully outline the defect area and reduce missed detections and false detections.

[0047] like Figure 5 As shown, step S40 specifically includes: S41: Control the visible light multi - band imager to perform area scanning on the surface of the photovoltaic panel, obtain multi - band spectral images covering a preset spectral range, where each pixel corresponds to a local area on the surface of the photovoltaic panel, and record the spectral data of the local area at different bands; Horizontally fix the photovoltaic panel to be detected on the automatic detection platform, and select a visible light multi - band imager. The spectral range of this device covers 400 - 1000 nm, the spectral resolution reaches 5 nm, and the spatial resolution is 0.5 mm / pixel. Before detection, turn on the standard D65 light source (color temperature 6500 K, color rendering index Ra≥90) to evenly illuminate the photovoltaic panel, and control the light intensity at 1000 ± 50 lux. Adjust the position of the imager so that its optical axis is perpendicular to the surface of the photovoltaic panel, and keep the distance at 1.2 meters to ensure that the field of view completely covers the photovoltaic panel (a 2m×1m - sized photovoltaic panel). Set the imager parameters as follows: scanning speed 0.1 m / s, exposure time 120 ms, select the band to cover the full spectral range of 400 - 1000 nm, and collect spectral images of 121 bands in total. Adopt the progressive scanning mode, and set the overlapping rate of adjacent scanning lines to 15% to avoid detection blind spots. During the scanning process, the imager automatically records the spectral data of each pixel at different bands. The data format is 16 - bit unsigned integer, and it is stored as a standard format file with geographic coordinate information.

[0048] S42: Pre - process the obtained multi - band spectral images to obtain pre - processed spectral images; First, perform radiometric calibration operations. Use the reference data collected by the standard diffuse reflection plate supporting the imager to convert the original DN (Digital Number) value of the image into actual reflectance data, eliminating the errors caused by differences in light intensity and device response. Then, adopt the median filtering algorithm to perform noise reduction on the image, set a 3×3 filtering window to remove the randomly distributed salt - and - pepper noise in the image while retaining the spectral feature details. Then, for the image with strip noise, utilize the correlation of adjacent bands to repair the strips. By calculating the mean difference of adjacent bands, interpolate and correct the abnormal strips. Finally, perform geometric correction on the processed image. Based on the preset control point coordinates on the surface of the photovoltaic panel, use the bilinear interpolation method to map the image pixels to the correct geographic coordinate positions, and output a pre - processed spectral image with a unified resolution of 1000×1000 pixels for subsequent spectral feature extraction.

[0049] S43: Adopt the spectral unmixing method to extract spectral feature data from the pre - processed spectral images; Spectral unmixing is performed using the Linear Mixing Model (LMM). First, according to the material characteristics of the photovoltaic panel and common defect types, 6 endmember spectra are selected: the photovoltaic panel substrate, glass cover plate, EVA film, crack, stain, and bubble, to construct an endmember library. For the preprocessed spectral image, the spectral curve of each pixel is regarded as a linear combination of the endmember spectra, and the mixing coefficients are solved by the non - negative least squares (NNLS) method. The formula is as follows: R ( λ ) = ∑ a i E i ( λ ) + ϵ ( λ ), where, R ( λ ) is the pixel spectrum, E i ( λ ) is the i th endmember spectrum, a i is the mixing coefficient, ϵ ( λ ) is the residual, and the value range of i is from 1 to n. After solving, the endmember abundance matrix of each pixel is obtained, and 12 spectral feature data such as the abundance value of each endmember, Spectral Angle Mapper (SAM) distance, and Spectral Information Divergence (SID) are extracted to form the feature vector of each pixel.

[0050] S44: Compare the spectral feature data with the pre - established defect spectral feature library of photovoltaic panels pixel by pixel. When the difference between the spectral feature data and the defect spectral features in the defect spectral feature library exceeds the set threshold, the corresponding area is determined as the spectral feature abnormal area.

[0051] Compare the extracted spectral feature data with the pre - established defect spectral feature library pixel by pixel. The defect spectral feature library contains 2000 groups of manually annotated standard defect spectral data, covering defect samples of different types and severities. During the comparison process, the Spectral Angle Mapper (SAM) algorithm is used to calculate the angle between the pixel spectrum and the defect spectra in the library. The formula is: . represents the angle between the pixel spectrum and the defect spectra in the library, which is used to measure the similarity of the spectral shapes of the two; is the spectral vector of the pixel to be detected, which is composed of the spectral data of each band of the pixel; is the spectral vector of the standard defect in the defect spectral feature library, containing the data of each band of the labeled defect samples; is the dot product of the two vectors, that is, multiply the corresponding band data and then sum; and are the norms of the spectral vector to be detected and the standard defect spectral vector respectively.

[0052] When the spectral angle θ is greater than the set threshold (set to 0.15 radians in this embodiment), and the spectral information divergence (SID) is greater than 2.0, it is determined that the area corresponding to the pixel is a spectral feature abnormal area. For the determined abnormal pixels, the 8-neighborhood connectivity algorithm is used for clustering, adjacent abnormal pixels are merged to form a complete abnormal area, and geometric parameters such as the area, centroid coordinates, and shape factor of each area are calculated. Finally, a visual detection result map with defect labels and a detailed detection report are generated.

[0053] The advantages of the above technical solution are as follows: Multi-band spectral image acquisition is adopted to cover the preset spectral range. Each pixel records the spectral data in different bands, which can capture the subtle spectral feature differences on the surface of the photovoltaic panel and provide rich information for defect detection; Noise and errors are eliminated through preprocessing to improve the image quality; The spectral unmixing method is used to extract spectral feature data, which can effectively separate the spectral contributions of different material components; By comparing with the pre-established defect spectral feature library pixel by pixel, various defects can be accurately identified, and the set difference threshold can flexibly adjust the detection sensitivity to meet the detection requirements of different types of photovoltaic panels, improving the accuracy and reliability of the detection.

[0054] Step S43 can specifically include: S431: Calculate the mean and standard deviation of the spectral data of each band, and determine the initial unmixing threshold according to the mean and the standard deviation; S432: Based on the spectral similarity of adjacent pixels, iteratively adjust the initial unmixing threshold through a sliding window algorithm to form a dynamic unmixing threshold; S433: According to the dynamic unmixing threshold, use a non-negative constrained linear spectral unmixing model to decompose the preprocessed spectral image into spectral feature data of different material components.

[0055] In step S431, for the preprocessed multi - band spectral image, statistical features are calculated along the band dimension. For each band, the spectral values of all pixels in the image are traversed. The arithmetic mean is calculated as the mean μ, and at the same time, the standard deviation σ is calculated to quantify the data dispersion. For example, at a specific band, the mean is obtained by summing the spectral values of 1000×1000 pixels and dividing by the total number, and then the data fluctuation range is calculated according to the standard deviation formula. Based on these statistical values, the initial unmixing threshold is set as T0 = μ + 1.5σ. The principle of this threshold setting is that the mean represents the average level of the spectral data of this band, while 1.5 times the standard deviation excludes the extreme values caused by noise while retaining most of the effective signals, thus initially distinguishing the meaningful spectral features from the background noise and providing a basic threshold reference for subsequent unmixing processing.

[0056] In step S432, the sliding window algorithm is used to perform spatial adaptive optimization on the initial threshold. A window with a size of 5×5 pixels slides pixel - by - pixel on the spectral image. For each central pixel in the window, the spectral similarity between it and the other 24 pixels in the window is calculated. The specific calculation method is as follows: the spectral data of the pixel is regarded as a high - dimensional vector, and the cosine value of the angle between the vectors is calculated to measure the spectral similarity. The smaller the angle, the more similar the spectra. If more than 70% of the pixels in the window have a spectral angle less than 0.1 radians (corresponding to a high similarity) with the central pixel, it is considered that the spectral consistency of this area is high, and the threshold corresponding to this window is reduced by 10% to enhance the sensitivity to fine spectral differences; conversely, if more than 30% of the pixels have a spectral angle greater than 0.2 radians (corresponding to a low similarity) with the central pixel, the threshold is increased by 10% to enhance the robustness to noise. Through this iterative adjustment of each window, a dynamic unmixing threshold matrix that changes with spatial position is formed. This matrix can automatically adjust the threshold according to local spectral characteristics, better adapting to the spectral complexity differences in different regions of the image while maintaining the overall unmixing accuracy.

[0057] In step S433, based on the dynamic threshold matrix, non - negative constrained linear spectral unmixing is performed for each pixel. The spectral response of a pixel is regarded as being composed of multiple material components (endmembers) mixed in different proportions, and each endmember corresponds to the characteristic spectrum of a specific material (such as glass, silicon substrate, EVA film, crack or contamination). By establishing a system of linear equations, the pixel spectrum is expressed as a linear combination of the endmember spectra, where the combination coefficients are the abundance values of the respective materials. During the solution process, two constraints are imposed: one is that all abundance values are non - negative (which conforms to physical reality, that is, the material content cannot be negative), and the other is that the sum of all abundance values is 1 (indicating that the pixel is completely composed of these endmember materials). The non - negative least - squares method is used to iteratively solve this system of equations: first, all abundance values are initialized to 0, and then the abundance values are gradually adjusted through 200 iterations. In each iteration, the residual between the fitted spectrum and the actual pixel spectrum under the current abundance combination is calculated, and the abundance values are adjusted according to the comparison result between the residual and the dynamic threshold. When the residual is less than the dynamic threshold at the corresponding position, it is considered that the fitting reaches a satisfactory accuracy and the iteration stops. Finally, the abundance matrix of each material component in each pixel is output, and this matrix precisely quantifies the spatial distribution of different materials on the surface of the photovoltaic panel.

[0058] As Figure 6 shown, step S50 specifically includes: S51: Control the excitation coil of the eddy current detection device to perform a row - by - row scan of the surface of the photovoltaic panel with a preset multi - frequency excitation signal. The multi - frequency excitation signal contains at least three different frequency components, and the amplitudes and phases of each frequency component are pre - configured according to the material characteristics of the photovoltaic panel; Fix the photovoltaic panel to be detected on the detection platform, and select an absolute - type coil probe with an outer diameter of 10 mm for the multi - frequency eddy current detector. According to the material characteristics of the photovoltaic panel (the thickness of the silicon substrate is 0.2 mm and the resistivity is about 10 Ω·cm), a three - frequency excitation signal is configured: a low - frequency component of 10 kHz (penetration depth is about 2.5 mm, used to detect deep - layer defects), a medium - frequency component of 50 kHz (penetration depth is about 1.1 mm, used to detect middle - layer defects), and a high - frequency component of 200 kHz (penetration depth is about 0.5 mm, used to detect surface defects). The amplitude ratio of each frequency component is set to 1:0.8:0.6, and the phase differences are set to 0°, 90°, and 180° to enhance the response differences of defects at different depths. Control the mechanical scanning platform to drive the probe to scan the surface of the photovoltaic panel row - by - row at a speed of 0.05 m / s, and the scanning pitch is 1 mm to ensure the detection coverage rate. During the scanning process, the detector generates a stable multi - frequency excitation signal through DDS (direct digital synthesis) technology, and drives the excitation coil through a power amplifier to ensure that the output power fluctuation is less than ±1%.

[0059] S52: Collect eddy current response signals corresponding to each excitation frequency through a receiving coil, perform a fast Fourier transform on the collected eddy current response signals, and obtain complex impedance data at different frequencies; The receiving coil synchronously collects eddy current response signals at each excitation frequency. The sampling frequency is set to 1 MHz, and the duration of each sampling is 10 ms to ensure that at least 10 data points are collected for each excitation period. After the collected time-domain signals are converted into digital signals through a 24-bit ADC (Analog-to-Digital Converter), they are transmitted to an embedded processor for real-time processing. Perform a fast Fourier transform (Fast Fourier Transform, FFT) on the time-domain signals at each sampling point to convert them into the frequency domain. Specifically, when implementing, the radix-2 FFT algorithm is used to process 1024-point data and calculate the complex impedance values of each frequency component (Z = R + jX, where R is the resistance component and X is the reactance component). To improve the signal-to-noise ratio, the FFT results at each detection point are averaged 5 times, and finally, complex impedance data at three frequencies of 10 kHz, 50 kHz, and 200 kHz are obtained and stored in a three-dimensional matrix format.

[0060] S53: Adopt a multi-frequency adaptive weight fusion algorithm, dynamically adjust the weights according to the signal-to-noise ratios of the complex impedance data at each frequency, and fuse the complex impedance data at different frequencies into a comprehensive characteristic index; When dealing with complex impedance data at different frequencies, an adaptive weight fusion algorithm is adopted to dynamically allocate weights according to the quality of the signals at each frequency, so as to enhance the expression ability of defect features. The specific implementation process is as follows: For the complex impedance data at each frequency (10 kHz, 50 kHz, 200 kHz), calculate its signal-to-noise ratio (SNR) respectively. The signal power is obtained by calculating the variance of the impedance data at all detection points at this frequency, which reflects the overall fluctuation intensity of the signal; the noise power is estimated by analyzing the impedance fluctuations in adjacent known defect-free regions, representing the background noise level. For example, at 10 kHz, if the impedance variance in the detection region is large while the fluctuation in the defect-free region is small, the SNR at this frequency is high, indicating that it is more sensitive to defects. Based on the SNR values of each frequency, weights are allocated proportionally: divide the SNR of each frequency by the sum of the SNR of all frequencies to obtain the normalized weight. For example, if the SNRs of 10 kHz, 50 kHz, and 200 kHz are 8, 5, and 3 respectively, their weights are 0.5, 0.3125, and 0.1875 in turn. This method ensures that frequencies with higher signal-to-noise ratios occupy a larger proportion in the final fusion, improving the reliability of defect detection. For each detection point, calculate the modulus difference between the impedance value at each frequency and the defect-free reference value (i.e., the absolute value of the impedance change), then multiply it by the corresponding frequency weight and sum to obtain the comprehensive feature index. For example, if the impedance changes at a certain point at three frequencies are 0.2 Ω, 0.15 Ω, and 0.1 Ω, and the corresponding weights are 0.5, 0.3125, and 0.1875, then its comprehensive index is 0.2×0.5 + 0.15×0.3125 + 0.1×0.1875 = 0.153 Ω.

[0061] S54: Compare the comprehensive feature index with a preset threshold. When the comprehensive feature index exceeds the preset threshold, mark the corresponding detection point as a suspected defect point; Compare the calculated comprehensive feature index with a preset threshold (set to 3.5 times the standard deviation in this embodiment). The standard deviation is obtained by statistically analyzing the comprehensive feature indexes of the known defect-free regions. When the comprehensive feature index of a certain detection point exceeds the threshold, mark this point as a suspected defect point and record its coordinate position. To reduce false positives, perform a secondary verification on the marked suspected defect points: check whether the impedance changes at at least two frequencies exceed their respective local thresholds (2 times the standard deviation of the impedance values in the normal regions at each frequency), and only the points that meet the conditions are retained as valid suspected defect points.

[0062] S55: Perform spatial clustering on all suspected defect points, merge the interconnected suspected defect points into the same region, and determine the abnormal region of the eddy current signal.

[0063] Perform spatial clustering analysis on all valid suspected defect points, and use the 8-neighborhood connectivity rule and region growing algorithm for merging. The specific steps are as follows: Step 1, select an untreated suspected defect point as the seed point; Step 2, check all points within its 8-neighborhood. If a point is also a suspected defect point and the distance from the seed point is less than 2 mm, add it to the current region; Step 3, repeat Step 2 for the newly added points until no further expansion is possible; mark the current region as a complete eddy current signal abnormal region, and record its geometric parameters such as boundary coordinates, area, perimeter, etc.; repeat Steps 1 to 4 until all suspected defect points are processed.

[0064] Finally, generate a detection result map containing all abnormal regions. Different colors in the map distinguish different types of defects (such as cracks, delaminations, pores, etc.), and perform preliminary classification through preset classification rules (based on the shape factor of the abnormal region, impedance change pattern, etc.).

[0065] Such as Figure 7 shown, Step S60 can specifically include: S61: Standardize the original coordinate data of the temperature abnormal region, deformation abnormal region, ultrasonic echo abnormal region, spectral feature abnormal region, and eddy current signal abnormal region to obtain the standardized coordinate data of each abnormal region; Import all the original coordinate data of the temperature abnormal region obtained by infrared thermal imaging detection, the deformation abnormal region obtained by laser speckle interferometry, the echo abnormal region identified by ultrasonic flaw detection, the spectral feature abnormal region determined by multi-band spectral analysis, and the signal abnormal region detected by eddy current testing into the data processing platform. Since the coordinate systems of different detection devices are different, for example, the infrared thermal imager measures positions in pixels and the ultrasonic flaw detector uses millimeters as the unit, unified processing is required. Using the min-max standardization method, adjust the horizontal and vertical coordinate data of all abnormal regions to the numerical range of 0 to 1. The specific operation is to first find the minimum and maximum values in all coordinate data. For each coordinate value, subtract the minimum value from it and then divide by the difference between the maximum value and the minimum value to obtain the standardized coordinate. Taking the temperature abnormal region as an example, if the minimum value of the original horizontal coordinate is 100 pixels, the maximum value is 800 pixels, and the original horizontal coordinate of a certain point is 300 pixels, then the standardized horizontal coordinate of this point is (300 - 100) ÷ (800 - 100), approximately equal to 0.286. After such processing, the coordinate data of abnormal regions obtained by different detection methods all have the same measurement scale, which is convenient for subsequent fusion analysis.

[0066] S62: Calculate the geometric feature parameters of each abnormal region respectively, and assign corresponding weights to the geometric feature parameters of each abnormal region according to the preset weight assignment rules; For each abnormal area that has completed standardization processing, a series of geometric feature parameters are calculated. These parameters include: area of the region, that is, the number of pixels or the actual area size contained in the region; perimeter, which is the total length of the region boundary; centroid coordinates, representing the geometric center position of the region; major axis length and minor axis length, used to describe the shape size of the region; circularity, by comparing the relationship between the area and perimeter of the region, to judge the degree of closeness of the region shape to a circle. The closer the value is to 1, the more circular the shape is. For example, for a certain ultrasonic echo abnormal area, after calculation, the area is 50 square millimeters and the perimeter is 30 millimeters, and its circularity is calculated to be approximately 0.70. According to the pre-set rules, weights are assigned to these geometric feature parameters. Since the area size directly reflects the severity of the defect, a higher weight of 0.4 is given to the area parameter; the perimeter can reflect the complexity of the defect boundary, and the weight is set to 0.2; the centroid coordinates are used to determine the defect position, with a weight of 0.1; the major and minor axis lengths are helpful for judging the defect shape, and the weights are both 0.1; circularity can assist in distinguishing defect types, and the weight is set to 0.1.

[0067] S63: According to the standardized coordinate data, the geometric feature parameters, and the weights corresponding to the geometric feature parameters, determine the overlapping parts of multiple abnormal areas through intersection operation, and mark the overlapping parts as initial defect candidate areas; Using a GIS (Geographic Information System) spatial analysis tool, such as ArcGIS software, or open-source libraries such as GDAL, based on the standardized coordinate data and the calculated geometric feature parameters, perform intersection operations on the five abnormal areas of temperature, deformation, ultrasonic echo, spectral features, and eddy current signals. These abnormal areas are regarded as polygon layers, and the overlapping parts are found through polygon overlay analysis. Specifically, when operating, each pixel or unit area within each area is judged one by one. If a certain area exists in two or more abnormal areas at the same time, this area is marked as the overlapping part. For example, if there is an area that appears in the temperature abnormal area, the ultrasonic echo abnormal area, and the spectral feature abnormal area at the same time, then this area will be included in the initial defect candidate area. During the process of judging the overlapping area, the weights assigned to each geometric feature parameter before are also combined to perform a weighted evaluation on the credibility of the overlapping part. The feature parameter with a higher weight has a greater influence when judging the overlapping area. After the weighted evaluation, the determined overlapping area becomes the initial defect candidate area, providing a basis for further accurately determining the defect position.

[0068] S64: Perform morphological filtering processing on the initial defect candidate area to obtain the final defect position.

[0069] For the initial defect candidate regions, a morphological filtering algorithm is used for optimization. First, an erosion operation is performed. A square template with a size of 3×3 is selected and slid successively over the candidate region. If all pixels within the region covered by the template belong to the candidate region, the pixel at the center of the template is retained; otherwise, the pixel is deleted. In this way, isolated pixel points and fine burrs on the boundary of the candidate region are removed, and tiny interference regions are eliminated. After the erosion operation is completed, a dilation operation is then carried out. Again, a 3×3 square template is used. The center of the template is aligned with the pixel of the candidate region. If there is at least one pixel belonging to the candidate region within the template, the pixel at the center of the template is marked as part of the candidate region, so as to restore the region shrunk due to erosion and connect adjacent small regions. After multiple repetitions of the erosion and dilation operations (in this embodiment, the loop is performed 3 times), discrete noise points and discontinuous regions are gradually removed, making the defect boundary smoother. Finally, a defect region with clear contours and accurate positions is obtained and output in the form of polygon vector data, which contains attributes such as detailed coordinate information, area size, and shape characteristics, and can be directly applied to the repair work and quality assessment of photovoltaic panel defects.

[0070] This step precisely integrates data from different detection methods (temperature, deformation, ultrasonic, spectral, and eddy current signals) through standardization processing and the application of weight assignment rules, ensuring that the characteristics of each abnormal region are reasonably quantified and fused. Overlapping regions are identified through intersection operations, further improving the accuracy of the defect position, while morphological filtering processing can effectively remove noise and misjudgments, ensuring the precise identification of the final defect position.

[0071] As Figure 8 shown, step S70 specifically includes: S71: Obtain temperature characteristics, deformation characteristics, echo characteristics, spectral characteristics, and eddy current characteristics respectively from the surface temperature distribution image, surface deformation data, ultrasonic echo data, spectral feature data, and eddy current signal feature data corresponding to the defect position; For the temperature characteristics, in the infrared thermal image, parameters such as the highest temperature, average temperature, temperature gradient (the ratio of the maximum temperature difference within the region to the distance), and the proportion of the hot spot area in the defect region are extracted. For example, the highest temperature of a certain defect region is 65°C, the average temperature is 58°C, and the temperature gradient is 0.8°C / mm, indicating local overheating abnormalities.

[0072] For the deformation characteristics, from the laser speckle interferometry data, the maximum deformation amount, average deformation amount, and deformation direction consistency index (determine the main deformation axis direction through principal component analysis) of the defect region are calculated. For example, the maximum deformation amount of a certain region is 0.15mm, the average deformation amount is 0.08mm, and the deformation direction is concentrated on the X-axis, suggesting the possible existence of linear cracks.

[0073] For the echo characteristics, wavelet transform is performed on the ultrasonic echo data to extract the energy distribution, peak frequency, and time-domain characteristics (such as echo time delay, attenuation coefficient). For example, if the echo energy in a certain area is reduced by 30% compared to the normal area and the peak frequency is shifted by 20 kHz, it indicates that there may be a delamination defect inside.

[0074] For the spectral characteristics, the reflectance ratio of the characteristic band (such as 650 nm / 800 nm), the offset of the absorption peak position, and the spectral angular distance are extracted from the multi-band spectral image. For example, if the reflectance at 650 nm in a certain area is significantly reduced and the absorption peak is shifted 5 nm towards the long-wave direction, it conforms to the material oxidation characteristics.

[0075] For the eddy current characteristics, principal component analysis is performed on the complex impedance data to extract the contribution rates of the first three principal components, the ratio of the real part to the imaginary part of the impedance, and the frequency response characteristics. For example, if the real part of the impedance in a certain area increases by 25% and the imaginary part decreases by 15% at a frequency of 10 kHz, it conforms to the metal material defect characteristics.

[0076] S72: According to the weight assignment model, determine the weights corresponding to the temperature characteristics, the deformation characteristics, the echo characteristics, the spectral characteristics, and the eddy current characteristics respectively; The weight assignment model is constructed using the Analytic Hierarchy Process (AHP), which includes the following steps: In the step of constructing the judgment matrix, according to expert experience or historical data, determine the influence degree of different characteristics on the defect type. For example, for crack defects, the weight of the deformation characteristics is 0.35, the weight of the echo characteristics is 0.3, the weight of the temperature characteristics is 0.15, the weight of the spectral characteristics is 0.1, and the weight of the eddy current characteristics is 0.1. In the consistency test step, calculate the maximum eigenvalue and the consistency index of the judgment matrix to ensure reasonable weight assignment. For example, when the consistency ratio CR is less than 0.1, the judgment matrix is considered to have satisfactory consistency. In the dynamic adjustment step, dynamically adjust the weights according to the depth information of the defect position. For example, for surface defects, the weight of the spectral characteristics is increased to 0.25; for internal defects, the weight of the echo characteristics is increased to 0.4.

[0077] S73: Perform weighted fusion on the temperature characteristics, the deformation characteristics, the echo characteristics, the spectral characteristics, the eddy current characteristics, and their respective corresponding weights to obtain a comprehensive feature vector; Before performing weighted fusion, standardize the temperature feature, deformation feature, echo feature, spectral feature, and eddy current feature to eliminate the numerical differences caused by different dimensions of each feature. For the temperature feature, subtract the lowest temperature in all detection data from the original temperature value, and then divide by the range of the temperature data to map it to the [0, 1] interval; for the deformation feature (in millimeters), normalize it based on the maximum deformation within the detection range by the method of "feature value / maximum deformation"; for the energy value in the echo feature, the reflectivity in the spectral feature, the impedance value in the eddy current feature, etc., all use a similar minimum-maximum normalization method. After completing the standardization, perform weighted calculation on each feature according to the weights determined by the weight distribution model in step S72. Assume that the weight of the temperature feature is 0.2, the weight of the deformation feature is 0.3, the weight of the echo feature is 0.25, the weight of the spectral feature is 0.1, the weight of the eddy current feature is 0.1, and the normalized temperature feature value is 0.7, the deformation feature value is 0.6, the echo feature value is 0.8, the spectral feature value is 0.4, and the eddy current feature value is 0.5. At this time, the weighted value of the temperature feature is 0.7×0.2 = 0.14, the weighted value of the deformation feature is 0.6×0.3 = 0.18, the weighted value of the echo feature is 0.8×0.25 = 0.2, the weighted value of the spectral feature is 0.4×0.15 = 0.06, and the weighted value of the eddy current feature is 0.5×0.1 = 0.05. Finally, arrange the weighted values of each feature in the order of temperature, deformation, echo, spectral, and eddy current to form a five-dimensional vector [0.14, 0.18, 0.2, 0.06, 0.05], that is, obtain the comprehensive feature vector. Through scientific normalization and weighted calculation, this vector not only retains the differences in the importance of features obtained by different detection methods but also unifies the multi-source heterogeneous data into a standard format.

[0078] S74: Analyze the comprehensive feature vector using a defect type recognition model based on deep learning to determine the type of surface defect of the photovoltaic panel; Adopt an optimized Convolutional Neural Network (CNN) as the defect type recognition model. The input layer of the model receives the comprehensive feature vector obtained in step S73. To adapt to the network structure, reshape the vector into a three-dimensional tensor form of 1×5×1, where 1 represents the number of samples, 5 corresponds to five feature dimensions, and 1 is the number of channels.

[0079] The main part of the network consists of 3 convolutional layers, 2 pooling layers, and 2 fully-connected layers. The first convolutional layer is set with 16 convolutional kernels of size 1×3, with a stride of 1, to extract features from the input tensor and explore the correlation patterns between different features. Subsequently, it is connected to a max-pooling layer with a pooling window of 1×2 and a stride of 2 to reduce the data dimension and computational complexity. The second convolutional layer is configured with 32 convolutional kernels of 1×3 to further extract high-order features. Then, it is connected to an average-pooling layer again, with a pooling window of 1×2 and a stride of 2. After two layers of convolution and pooling operations, the output is flattened into a one-dimensional vector, which is successively connected to fully-connected layers containing 128 neurons and 64 neurons. The ReLU activation function is used between the fully-connected layers to enhance the non-linear expression ability of the network. The final output layer is set with 5 neurons, corresponding to 5 common defect types of photovoltaic panels (cracks, delamination, bubbles, stains, oxidation), and the Softmax activation function is used to output the probability values corresponding to each category.

[0080] During the model training phase, a dataset containing 20,000 labeled samples is used, among which 16,000 samples are used as the training set, 2,000 samples are used as the validation set, and 2,000 samples are used as the test set. The Adam optimizer is adopted, with the initial learning rate set to 0.001, and the learning rate decays by 0.9 every 10 training epochs. The cross-entropy loss function is selected as the loss function, and the network parameters are updated through the backpropagation algorithm during the training process, with a total of 50 training epochs. When the loss of the validation set no longer decreases and the accuracy tends to be stable, the training is stopped and the optimal model parameters are saved.

[0081] In actual application, the comprehensive feature vector to be detected is input into the trained model. Among the 5 output probability values, the category with the highest probability is the type of surface defect of the photovoltaic panel determined by the model. For example, if the output probability vector is [0.05, 0.85, 0.03, 0.04, 0.03], it is determined that there is a delamination defect on the surface of the photovoltaic panel, and the probability value output by the model can also be used as a reference for the confidence level of the determination result.

[0082] S75: Determine the size of the surface defect of the photovoltaic panel according to the temperature feature, the deformation feature, and the echo feature.

[0083] First, corresponding quantization methods are adopted for different feature data to obtain preliminary size information. For the temperature feature, the temperature difference between the defect area and the normal area in the infrared thermal image is used for definition. The temperature gradient threshold is set to 0.3℃ / mm, and the edge detection algorithm (Canny algorithm) is used to identify the boundary of the area where the temperature gradient exceeds the threshold, and the equivalent diameter of this area is calculated as the defect size based on the temperature feature. For example, the equivalent diameter of a certain defect area is calculated to be 8mm.

[0084] For the deformation feature, based on the surface displacement data obtained by laser speckle interferometry, using the amount of deformation as the judgment basis, the area where the amount of deformation exceeds 0.03 mm is regarded as the defective area. The contour tracking algorithm is used to extract the area contour, and then the area enclosed by the contour is calculated and converted into an equivalent diameter. Suppose the equivalent diameter calculated for a certain area is 6 mm.

[0085] For the echo feature, analyze the echo data collected by ultrasonic flaw detection. Determine the defect range through the attenuation degree and time delay of the echo signal. When the energy attenuation of the echo signal exceeds 40% and the time delay exceeds the normal range (set as ±0.5 μs), mark the corresponding area as the defective area. Use the threshold segmentation algorithm to extract the defective area and calculate its longest straight-line distance as the defect size based on the echo feature. For example, the measured value of a certain area is 7 mm.

[0086] Finally, according to the different types of defects, adopt a dynamic weight allocation strategy to fuse the above three size data. Set the weights through historical data statistics and expert experience: for crack defects, the weight of the deformation feature is 0.4, the weight of the temperature feature is 0.3, and the weight of the echo feature is 0.3; for delamination defects, the weight of the echo feature is 0.5, the weight of the temperature feature is 0.3, and the weight of the deformation feature is 0.2. Multiply the size corresponding to each feature by the weight and sum them to obtain the final defect size. For example, for a certain crack defect, the final calculated defect size is: 8×0.3 + 6×0.4 + 7×0.3 = 7.2 mm, which is used as the accurate judgment result of the surface defect size of the photovoltaic panel.

[0087] As Figure 9 shown, this embodiment provides a photovoltaic panel surface defect detection system based on physical property analysis. The system is used to execute any one of the foregoing methods. The system includes: An infrared thermal imaging detection module, which is used to perform a full-field scan of the photovoltaic panel using an infrared thermal imager, obtain the surface temperature distribution image of the photovoltaic panel, and determine the temperature abnormal area based on the surface temperature distribution image; A laser speckle deformation detection module, which is used to scan the surface of the photovoltaic panel using a laser speckle interferometer, obtain the surface deformation data of the photovoltaic panel, and determine the deformation abnormal area based on the surface deformation data; An ultrasonic flaw detection module, which is used to perform an ultrasonic scan of the surface of the photovoltaic panel using an ultrasonic flaw detector, obtain the ultrasonic echo data on the surface of the photovoltaic panel, and determine the ultrasonic echo abnormal area based on the ultrasonic echo data; A spectral feature analysis module, which is used to obtain the spectral image of the surface of the photovoltaic panel using a visible light multi-band imager, extract spectral feature data from the spectral image based on the spectral unmixing algorithm, and determine the spectral feature abnormal area according to the spectral feature data; Eddy current signal detection module, which is used to detect the surface area of the photovoltaic panel by using eddy current detection equipment, obtain eddy current signal characteristic data, and determine the eddy current signal abnormal area based on the eddy current signal characteristic data; Defect position determination module, which is used to determine the defect position according to the temperature abnormal area, the deformation abnormal area, the ultrasonic echo abnormal area, the spectral characteristic abnormal area and the eddy current signal abnormal area; Defect identification and analysis module, which is used to determine the photovoltaic panel surface defect identification result including the type and size of the surface defect according to the surface temperature distribution image, the surface deformation data, the ultrasonic echo data, the spectral characteristic data and the eddy current signal characteristic data corresponding to the defect position.

[0088] Those skilled in the art can clearly understand that, for the convenience and conciseness of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. The specific working process of the units and modules in the above system can refer to the corresponding process in the foregoing method embodiments and will not be elaborated here.

[0089] The embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements any one of the above methods.

[0090] Next, refer to Figure 10 which shows a schematic structural diagram of a computer system 800 of an electronic device suitable for implementing the embodiment of the present invention. Figure 10 The electronic device shown is only an example and should not bring any limitation to the functions and usage scope of the embodiment of the present invention. As Figure 10 shown, the computer system 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 802 or the program loaded from the storage part 808 into the random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the computer system 800 are also stored. The CPU 801, the ROM 802 and the RAM 803 are connected to each other through a bus 804. The input / output (I / O) interface 805 is also connected to the bus 804.

[0091] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, etc.; an output section 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as required. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is mounted on the drive 810 as required so that a computer program read therefrom is installed into the storage section 808 as required.

[0092] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting surface defects of a photovoltaic panel based on physical property analysis, characterized in that, Including the following steps: S10: Use an infrared thermal imager to perform a full-field scan of the photovoltaic panel, obtain the surface temperature distribution image of the photovoltaic panel, and determine the temperature anomaly area based on the surface temperature distribution image; S20: Use a laser speckle interferometer to scan the surface of the photovoltaic panel, obtain the surface deformation data of the photovoltaic panel, and determine the deformation anomaly area based on the surface deformation data; S30: Use an ultrasonic flaw detector to perform an ultrasonic scan of the surface of the photovoltaic panel, obtain the ultrasonic echo data of the surface of the photovoltaic panel, and determine the ultrasonic echo anomaly area based on the ultrasonic echo data; S40: Use a visible light multi-band imager to obtain the spectral image of the surface of the photovoltaic panel, extract spectral feature data from the spectral image based on the spectral unmixing algorithm, and determine the spectral feature anomaly area according to the spectral feature data; S50: Use an eddy current detection device to detect the surface area of the photovoltaic panel, obtain eddy current signal feature data, and determine the eddy current signal anomaly area based on the eddy current signal feature data; S60: Determine the defect position according to the temperature anomaly area, the deformation anomaly area, the ultrasonic echo anomaly area, the spectral feature anomaly area, and the eddy current signal anomaly area; S70: Determine the photovoltaic panel surface defect identification result including the type and size of the surface defect according to the surface temperature distribution image, the surface deformation data, the ultrasonic echo data, the spectral feature data, and the eddy current signal feature data corresponding to the defect position.

2. The method according to claim 1, characterized in that Step S10 specifically includes: S11: Set the scanning range of the infrared thermal imager to cover the entire surface of the photovoltaic panel, perform a full-field scan of the photovoltaic panel, and obtain multiple frames of surface temperature distribution images; S12: Use the adaptive threshold filtering algorithm to perform noise reduction processing on the surface temperature distribution image to obtain the preprocessed surface temperature distribution image; S13: Calculate the temperature deviation between the temperature value of each pixel point in the preprocessed surface temperature distribution image and the preset normal temperature range, and mark the pixel points with the temperature deviation exceeding the dynamic threshold as abnormal points; S14: Perform connectivity analysis on the pixel points marked as abnormal points, use the connected region algorithm to classify adjacent abnormal points into the same abnormal sub-region, and generate a preliminary regional map of the temperature anomaly area; within each abnormal sub-region of the preliminary regional map, use the region expansion algorithm to further connect adjacent but unconnected abnormal points according to the preset spatial distance threshold to form a temperature anomaly area with continuous boundaries.

3. The method according to claim 1, characterized in that Step S20 specifically includes: S21: Control the laser speckle interferometer to scan the surface of the photovoltaic panel with pulsed laser at a preset frequency. During the scanning process, use the spatial light modulator in the laser speckle interferometer to perform real-time modulation on the laser speckle pattern, so that the laser speckle pattern forms a dynamically changing modulated speckle field on the surface of the photovoltaic panel; S22: Collect the modulated speckle image sequence corresponding to the modulated speckle field on the surface of the photovoltaic panel by the camera at a frame rate synchronized with the pulsed laser; S23: Process the sequence of modulated speckle images using a deep learning-based phase unwrapping algorithm to obtain the restored phase information, and extract the surface deformation data of the photovoltaic panel from the restored phase information; S24: Analyze the surface deformation data through a clustering analysis algorithm based on a preset deformation threshold to determine the deformation abnormal area.

4. The method according to claim 1, wherein Step S30 specifically includes: S31: Control the ultrasonic flaw detector to perform ultrasonic scanning row by row on the surface of the photovoltaic panel, and synchronously collect the ultrasonic echo data of each scanning point; S32: Input the collected ultrasonic echo data into a pre-trained convolutional neural network model to output the probability value of defects at each scanning point; S33: According to a preset probability threshold, determine the area where the scanning points with probability values higher than the preset probability threshold are located as the ultrasonic echo abnormal area.

5. The method according to claim 1, characterized in that, Step S40 specifically includes: S41: Control the visible light multi-band imager to perform area scanning on the surface of the photovoltaic panel to obtain a multi-band spectral image covering a preset spectral range. Each pixel corresponds to a local area on the surface of the photovoltaic panel, and record the spectral data of the local area in different bands; S42: Preprocess the obtained multi-band spectral image to obtain the preprocessed spectral image; S43: Use a spectral unmixing method to extract spectral feature data from the preprocessed spectral image; S44: Compare the spectral feature data pixel by pixel with a pre-established defect spectral feature library of the photovoltaic panel. When the difference between the spectral feature data and the defect spectral features in the defect spectral feature library exceeds a set threshold, determine the corresponding area as the spectral feature abnormal area.

6. The method according to claim 5, wherein Step S43 specifically includes: S431: Calculate the mean and standard deviation of the spectral data of each band, and determine the initial unmixing threshold according to the mean and the standard deviation; S432: Based on the spectral similarity of adjacent pixels, iteratively adjust the initial unmixing threshold through a sliding window algorithm to form a dynamic unmixing threshold; S433: According to the dynamic unmixing threshold, use a non-negative constrained linear spectral unmixing model to decompose the preprocessed spectral image into spectral feature data of different material components.

7. The method according to claim 1, wherein Step S50 specifically includes: S51: Control the excitation coil of the eddy current detection device to perform row-by-row scanning on the surface of the photovoltaic panel with a preset multi-frequency excitation signal; S52: Collect the eddy current response signals corresponding to each excitation frequency through the receiving coil, perform a fast Fourier transform on the collected eddy current response signals, and obtain the complex impedance data at different frequencies; S53: Use a multi-frequency adaptive weight fusion algorithm to dynamically adjust the weights according to the signal-to-noise ratio of the complex impedance data at each frequency, and fuse the complex impedance data at different frequencies into a comprehensive feature index; S54: Compare the comprehensive feature index with a preset threshold. When the comprehensive feature index exceeds the preset threshold, mark the corresponding detection point as a suspected defect point; S55: Perform spatial clustering on all suspected defect points, merge the connected suspected defect points into the same area, and determine the eddy current signal abnormal area.

8. The method according to claim 1, wherein Step S60 specifically includes: S61: Standardize the original coordinate data of the temperature anomaly region, deformation anomaly region, ultrasonic echo anomaly region, spectral feature anomaly region, and eddy current signal anomaly region to obtain the standardized coordinate data of each anomaly region; S62: Calculate the geometric feature parameters of each anomaly region respectively, and assign corresponding weights to the geometric feature parameters of each anomaly region according to the preset weight assignment rule; S63: Determine the overlapping part of multiple anomaly regions through intersection operation according to the standardized coordinate data, the geometric feature parameters, and the weights corresponding to the geometric feature parameters, and mark the overlapping part as the initial defect candidate region; S64: Perform morphological filtering processing on the initial defect candidate region to obtain the final defect position.

9. The method according to claim 1, characterized in that, Step S70 specifically includes: S71: Obtain the temperature feature, deformation feature, echo feature, spectral feature, and eddy current feature respectively from the surface temperature distribution image, surface deformation data, ultrasonic echo data, spectral feature data, and eddy current signal feature data corresponding to the defect position; S72: Determine the weights corresponding to the weights of the temperature feature, the deformation feature, the echo feature, the spectral feature, and the eddy current feature respectively according to the weight assignment model; S73: Perform weighted fusion on the temperature feature, the deformation feature, the echo feature, the spectral feature, and the eddy current feature and their respective corresponding weights to obtain a comprehensive feature vector; S74: Analyze the comprehensive feature vector by using a deep learning-based defect type recognition model to determine the type of the surface defect of the photovoltaic panel; S75: Determine the size of the surface defect of the photovoltaic panel according to the temperature feature, the deformation feature, and the echo feature.

10. A photovoltaic panel surface defect detection system based on physical property analysis, characterized in that, The system is used to execute the method described in any one of claims 1-9, and the system includes: An infrared thermal imaging detection module, which is used to perform a full-field scan of the photovoltaic panel by using an infrared thermal imager, obtain the surface temperature distribution image of the photovoltaic panel, and determine the temperature anomaly region based on the surface temperature distribution image; A laser speckle deformation detection module, which is used to scan the surface of the photovoltaic panel by using a laser speckle interferometer, obtain the surface deformation data of the photovoltaic panel, and determine the deformation anomaly region based on the surface deformation data; An ultrasonic flaw detection module, which is used to perform ultrasonic scanning on the surface of the photovoltaic panel by using an ultrasonic flaw detector, obtain the ultrasonic echo data on the surface of the photovoltaic panel, and determine the ultrasonic echo anomaly region based on the ultrasonic echo data; A spectral feature analysis module, which is used to obtain the spectral image of the surface of the photovoltaic panel by using a visible light multi-band imager, extract spectral feature data from the spectral image based on the spectral unmixing algorithm, and determine the spectral feature anomaly region according to the spectral feature data; An eddy current signal detection module, which is used to detect the surface area of the photovoltaic panel by using an eddy current detection device, obtain eddy current signal feature data, and determine the eddy current signal anomaly region based on the eddy current signal feature data; A defect location determination module, configured to determine the defect location according to the temperature anomaly region, the deformation anomaly region, the ultrasonic echo anomaly region, the spectral feature anomaly region, and the eddy current signal anomaly region; A defect identification and analysis module, configured to determine a photovoltaic panel surface defect identification result including the type and size of the surface defect according to the surface temperature distribution image, the surface deformation data, the ultrasonic echo data, the spectral feature data, and the eddy current signal feature data corresponding to the defect location.

Images