Photovoltaic module defect detection method and system

Abstract

This invention discloses a method and system for detecting defects in photovoltaic modules. The method includes applying thermal and optical excitation to the photovoltaic module and acquiring thermal and optical image sequences; filtering the thermal and optical image sequences to obtain clear thermal and optical image sets corresponding to the acquisition time; fusing features of thermal and optical images acquired at the same time to obtain a photothermal feature fused image set; labeling each image in the photothermal feature fused image set with corresponding attribute tags; training a defect detection model using the attribute-labeled images to obtain a target defect detection model; and determining the defect detection result corresponding to the photovoltaic module image when acquiring an image of the photovoltaic module to be detected using the target defect detection model. This invention overcomes the shortcomings of traditional technologies, such as difficulty in detecting small-area defects in photovoltaic modules and the limitation of single-mode detection images.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic module defect detection technology, and particularly relates to a method and system for photovoltaic module defect detection using photoluminescence technology assisted by electromagnetic induction thermal imaging. Background Technology

[0002] Internationally, U. Netzelmann et al. at the Fraunhofer Institute for Nondestructive Testing in Germany attempted to use electromagnetic induction thermal imaging to detect damage in crystalline photovoltaic cells. Domestically, Hunan University studied the thermal radiation phenomenon caused by electromagnetically induced planar eddy currents in crystalline photovoltaic cells and established a nondestructive testing method and system for photovoltaic cells based on induced photothermal radiation (patent authorization announcement number CN107192759B). First, electromagnetic induction excitation is applied to the photovoltaic cell through an electromagnetic induction excitation coil, causing the photovoltaic cell to generate light and thermal radiation. Then, a short-wave infrared camera and thermal imager are used to perform high-precision and high-sensitivity imaging of the light and thermal radiation. Planar array induction coils are used to collect multi-frequency impedance information from different regions of the photovoltaic cell and module. The multi-frequency impedance information, light radiation amplitude, temperature change amplitude, and temperature change phase of the photovoltaic cell are comprehensively used as feature values. Support vector machine algorithms or artificial neural network algorithms are used to process the multi-frequency impedance information, light radiation signals, and thermal radiation signals to extract feature values ​​that characterize attributes, defects, and health. The control system then determines whether the tested photovoltaic cell is faulty.

[0003] While defect detection using transient images of electrothermal and electroluminescence in bare photovoltaic cells is well-suited for quality control on photovoltaic production lines, it is not entirely applicable to the operation and maintenance management of finished modules. In finished modules, photovoltaic cells are encapsulated within EVA film and tempered glass, protected by these layers. This protective layer obscures the surface of the photovoltaic cells, hindering the propagation of light and heat radiation under electromagnetic induction. The light and heat radiation generated by the photovoltaic cells inside the module must pass through this protective layer to reach the module surface; this propagation process incurs losses, affecting the detection results. Figure 1 The specific structure of a certain crystalline silicon photovoltaic module is shown.

[0004] Early and small-area defects in photovoltaic (PV) modules mainly include scratches, microcracks, fractures, encapsulant film discoloration, hot spots, PID effects, and snail trails. These differ in type from defects in bare PV cells (cell breakage, hot spots, cracks, etc.). Furthermore, since it takes time for heat radiation from PV cells to propagate to the outer surface of the module glass, further analysis of the thermal image sequences during the electromagnetic induction heating process of PV modules can be conducted to explore the influence of defect type and size on the temporal characteristics of thermal imaging images. This allows for the capture of useful information from finished PV modules that differs from that found in bare PV cells. Research indicates that the study of temporal image defect detection methods for PV modules based on electromagnetic induction thermal imaging is still in its early stages, with limited relevant literature, but its development prospects are broad.

[0005] However, thermal image sequences acquired by thermal imagers exhibit lateral thermal diffusion, which may affect the accuracy of electromagnetic induction thermal imaging detection of defects. Photoluminescence detection technology, on the other hand, uses an external light source to illuminate photovoltaic modules, causing internal charge carrier transitions and generating light radiation. This light radiation is then used for imaging and defect detection, enabling the imaging and detection of defects within the cells of photovoltaic modules.

[0006] As shown in Table 1, existing photovoltaic module thermal imaging inspection technologies include handheld thermal imager inspection, ground robot-mounted thermal imager inspection, and UAV-borne thermal imager inspection. These technologies mainly rely on the heat generated when photovoltaic modules generate electricity to achieve detection. However, photovoltaic modules easily reach a thermal steady state when generating electricity, and the surface temperature of the photovoltaic module in the defect area tends to be the same as that in the normal area. The defect features in the thermal image are not obvious, making it difficult to detect early defects. The inspection method of UAV combined with thermal imaging is difficult to detect small-area defects with a scale of 10mm because the distance between the UAV and the inspected module is too far. Existing thermal imaging inspection technologies can only be used for inspection during the day when electricity is generated. They are easily affected by environmental factors such as solar radiation and ground reflection, resulting in poor imaging effects and wasting the best maintenance time at night when electricity is not generated.

[0007] Table 1 Summary of Existing Thermal Imaging Detection Technologies for Photovoltaic Modules

[0008] Summary of the Invention

[0009] The purpose of this invention is to provide a method and system for detecting defects in photovoltaic modules, in order to solve the problems that photovoltaic modules are prone to thermal steady state during power generation and it is difficult to detect their early defects, that UAVs combined with thermal imaging detection technology are difficult to detect small-area defects due to their low field of view resolution, and that traditional thermal imaging detection technology is only applicable during the day.

[0010] The present invention solves the above-mentioned technical problems through the following technical solution:

[0011] A method for detecting defects in photovoltaic modules, the method comprising the following steps:

[0012] Simultaneously apply thermal and optical excitation to the photovoltaic module, and simultaneously acquire thermal and optical image sequences of the photovoltaic module;

[0013] The thermal image sequence and the optical image sequence are filtered to obtain a clear set of thermal images and a set of optical images that correspond to the acquisition time.

[0014] By fusing features of thermal and optical images acquired at the same time, a set of optical-thermal feature fusion images is obtained.

[0015] Each image in the photothermal feature fusion image set is labeled with a corresponding attribute tag;

[0016] The defect detection model is trained using images with attribute labels to obtain the target defect detection model;

[0017] When an image of a photovoltaic module to be inspected is acquired, the target defect detection model is used to determine the defect detection result corresponding to the image of the photovoltaic module to be inspected.

[0018] Furthermore, before simultaneously applying thermal and light excitation to the photovoltaic module, the detection method further includes: determining optimal thermal excitation parameters and optimal lift-off distance, and setting a thermal excitation source according to the optimal thermal excitation parameters and optimal lift-off distance to apply thermal excitation to the photovoltaic module.

[0019] Furthermore, the specific implementation process for determining the optimal thermal excitation parameters and the optimal lift-off distance includes:

[0020] A coupled simulation model consisting of a thermal excitation source model and a photovoltaic module model is constructed, with the thermal excitation source model located directly above the photovoltaic module model;

[0021] Using the coupled simulation model, a first relationship curve between the temperature of the upper surface of the photovoltaic module model and the thermal excitation application time under different excitation currents is obtained, as well as a second relationship curve between the eddy current density on the surface of the cell inside the photovoltaic module model and the lift-off distance; wherein, the lift-off distance refers to the vertical distance between the thermal excitation source model and the upper surface of the photovoltaic module model;

[0022] The optimal thermal excitation parameters are the excitation current magnitude and thermal excitation application time corresponding to the first relationship curve that minimizes the time required to heat the upper surface of the photovoltaic module model from the first temperature to the second temperature; wherein, both the first temperature and the second temperature are within the operating temperature range of the photovoltaic module; and in the second relationship curve, the lift-off distance corresponding to the maximum eddy current density is the optimal lift-off distance.

[0023] Furthermore, the specific process for obtaining the first relationship curve is as follows:

[0024] By setting a lift-off distance and keeping it constant, changing the magnitude of the excitation current of the thermal excitation source model, and applying different thermal excitations to the photovoltaic module model, a first relationship curve is obtained showing the temperature of the upper surface of the photovoltaic module model changing with the thermal excitation application time under different excitation currents.

[0025] The specific process for obtaining the second relationship curve is as follows:

[0026] Set the thermal excitation parameters of the thermal excitation source model and keep the thermal excitation parameters unchanged, change the lift-off distance, and obtain the second relationship curve of the eddy current density on the surface of the cell inside the photovoltaic module model as a function of the lift-off distance;

[0027] The thermal excitation parameters are the optimal thermal excitation parameters.

[0028] Furthermore, before determining the optimal thermal excitation parameters based on the first relationship curve and the optimal lift-off distance based on the second relationship curve, the detection method further includes a step of verifying the first and second relationship curves. The specific implementation process is as follows:

[0029] The first experimental platform was constructed, consisting of a physical thermal excitation source, thermal image acquisition equipment, and a physical photovoltaic module.

[0030] Based on the first test platform, under the same thermal excitation parameters and lift-off distance as the first relationship curve, thermal excitation was applied to the photovoltaic module to obtain a first test curve showing the change of the temperature of the upper surface of the photovoltaic module with the thermal excitation application time under different excitation currents.

[0031] Based on the first test platform, under the same thermal excitation parameters and lift-off distance as the obtained second relationship curve, thermal excitation was applied to the photovoltaic module to obtain a second test curve showing the change of eddy current density on the surface of the cell inside the photovoltaic module with the lift-off distance.

[0032] The first relationship curve is compared with the first experimental curve. If the relative error between the first relationship curve and the first experimental curve is less than the error threshold, the first relationship curve is verified. Otherwise, the simulation is repeated to obtain the first relationship curve.

[0033] The second relationship curve is compared with the second experimental curve. If the relative error between the second relationship curve and the second experimental curve is less than the error threshold, the second relationship curve is verified. Otherwise, the simulation is repeated to obtain the second relationship curve.

[0034] Furthermore, before simultaneously applying thermal and optical excitation to the photovoltaic module, the detection method further includes: a step of determining the image extraction time period based on the optimal thermal excitation parameters and the optimal lift-off distance, the specific implementation of which includes:

[0035] A second experimental platform was constructed, consisting of a thermal excitation source, an external light source, photovoltaic modules, and optical image acquisition equipment; wherein the photovoltaic modules were defective.

[0036] The thermal excitation source of the second test platform is set according to the optimal thermal excitation parameters and the optimal lift-off distance;

[0037] The photovoltaic module is simultaneously subjected to thermal excitation and light excitation using the thermal excitation source and an external light source, and the light image sequence of the photovoltaic module is simultaneously acquired using a light image acquisition device.

[0038] The sharpness of each frame in the optical image sequence is evaluated to obtain a third relationship curve between the average gray value of the image and the thermal excitation application time;

[0039] If the optimal relationship curve is positively correlated with the third relationship curve, then the later thermal excitation application time period in the optimal relationship curve is selected as the image extraction time period.

[0040] If the optimal relationship curve is negatively correlated with the third relationship curve, then the earlier thermal excitation application time period in the optimal relationship curve is selected as the image extraction time period.

[0041] If the optimal relationship curve and the third relationship curve are partially positively correlated and partially negatively correlated, then the thermal excitation application time period corresponding to the positive correlation in the optimal relationship curve is selected as the image extraction time period.

[0042] Wherein, the optimal relationship curve is the first relationship curve corresponding to the optimal thermal excitation parameters determined by the coupled simulation model;

[0043] Before filtering the thermal image sequence and the optical image sequence, thermal image sequence and optical image sequence corresponding to the acquisition time and the image extraction time period are extracted from the thermal image sequence and the optical image sequence for filtering.

[0044] Furthermore, the Tenengrad gradient function is used to evaluate the sharpness of each frame in the optical image sequence. The specific implementation process is as follows:

[0045] The Sobel operator is used to extract the horizontal and vertical gradients for each pixel in each frame of the light image sequence;

[0046] The gradient value of each pixel is obtained by summing the squares of the horizontal and vertical gradients.

[0047] The average grayscale value of each frame of the image is calculated based on the gradient value of each pixel.

[0048] The average gray value of each frame in the light image sequence is calculated to obtain the third relationship curve.

[0049] Furthermore, feature fusion of thermal and optical images acquired at the same time is performed using a deep convolutional generative adversarial network (GAN) approach. The specific implementation process includes:

[0050] Alignment operations are performed on thermal and optical images acquired at the same time to achieve registration between thermal and optical images;

[0051] The registered thermal image and optical image acquired at the same time are input into the generator network G to obtain a single fused image. Then, the single fused image and the label image are input into the discriminator network D to obtain a high-quality optical-thermal feature fused image.

[0052] In this network, the generator network G connects the optical and thermal images using a concatenate function, extracts high-dimensional features from the images through residual unit blocks, and obtains a single fused image. The label image is generated by dividing the registered thermal and optical images acquired at the same time into sub-blocks using a sliding window, and then using a fusion rule that selects the larger gradient of the high-frequency part and performs a weighted average of the low-frequency part, and using Laplacian pyramid transformation. The discriminator network D uses stride-length convolution to extract the depth features of the single fused image and the label image respectively, and uses the LeakyReLU activation function to enhance the nonlinearity of the network. Finally, the optical and thermal feature fused image is output as a fully connected layer.

[0053] Based on the same concept, the present invention also provides a photovoltaic module defect detection system, the system comprising:

[0054] A thermal excitation source and a light excitation source are disposed above the photovoltaic module, and the thermal excitation source and the light excitation source are used to simultaneously apply thermal excitation and light excitation to the photovoltaic module;

[0055] Thermal image acquisition equipment is used to acquire a sequence of thermal images of photovoltaic modules when thermal excitation is applied to them;

[0056] An optical image acquisition device is used to acquire a sequence of optical images of a photovoltaic module when optical excitation is applied to it.

[0057] A data processing device is connected to the thermal image acquisition device and the optical image acquisition device respectively. The data processing device is used to acquire thermal image sequences and optical image sequences of photovoltaic modules; to filter the thermal image sequences and optical image sequences respectively to obtain clear thermal image sets and optical image sets corresponding to the acquisition time; to perform feature fusion on thermal images and optical images at the same acquisition time to obtain photothermal feature fusion images, and then to obtain a photothermal feature fusion image set; to label each image in the photothermal feature fusion image set with corresponding attribute labels; to train a defect detection model using the images with attribute labels to obtain a target defect detection model; and when acquiring an image of a photovoltaic module to be detected, to use the target defect detection model to determine the defect detection result corresponding to the image of the photovoltaic module to be detected.

[0058] Beneficial effects

[0059] Compared with the prior art, the advantages of the present invention are as follows:

[0060] This invention applies both light and heat excitation to a photovoltaic module simultaneously. Under these excitations, the module's temperature rises, generating a continuous sequence of light and heat radiation images. The light and heat radiation images are then fused to identify features, and the resulting images are used for defect detection. By combining the advantages of thermal imaging and short-wave infrared imaging, this invention expands the types of defects that can be detected in photovoltaic modules, improves the accuracy of defect detection, and enables the detection of early-stage and small-area defects. This overcomes the limitations of traditional thermal imaging detection techniques, which struggle to detect small-area defects in photovoltaic modules and suffer from limited image modalities. Attached Figure Description

[0061] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is a schematic diagram of the specific structure of a crystalline silicon photovoltaic module in the background art of this invention;

[0063] Figure 2 This is a flowchart of the photovoltaic module defect detection method in an embodiment of the present invention;

[0064] Figure 3 This is a diagram of the coupled simulation model in an embodiment of the present invention;

[0065] Figure 4 This is a schematic diagram of the photovoltaic module defect detection system in an embodiment of the present invention;

[0066] Figure 5This is a schematic diagram of the photovoltaic module defect detection system in an embodiment of the present invention.

[0067] Among them, 1-shortwave infrared camera, 2-external light source, 3-induction coil, 4-thermal imager, 5-photovoltaic module, 51-light radiation, 52-glass, 53-battery cell, 54-encapsulation film, 55-induced current, 56-defect, 57-thermal radiation, 58-backplate, 59-interconnect strip, 60-silicone, 61-junction box, 62-aluminum frame. Detailed Implementation

[0068] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] The technical solutions of this application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0070] Definitions:

[0071] Electromagnetic induction thermal imaging inspection: also known as eddy current thermal imaging inspection, uses an external electromagnetic induction excitation source to generate eddy currents inside the object under test. The Joule heat generated by the eddy currents propagates inside the object under test, affecting the temperature distribution on the surface of the object under test. By processing and analyzing the thermal images of the surface of the object under test acquired by the thermal imager, the internal defects of the object under test can be detected.

[0072] Lateral thermal diffusion phenomenon: When photovoltaic modules have scratches, cracks, or other defects, they will produce bright spots with obvious defect outlines on the thermal image due to the different thermal conductivity rates of air and solids (solids conduct heat faster and their temperature drops faster). However, as the photovoltaic module is continuously heated, the heat from the defective area will diffuse laterally to the surrounding area, thus blurring the defect outline of the bright spots.

[0073] Photoluminescence technology: Objects rely on external light sources to obtain energy, which excites them and causes them to emit light.

[0074] Temporal images: Image sequences acquired in chronological order according to a certain sampling frequency.

[0075] Semantic segmentation: A deep learning algorithm that associates labels or categories with each pixel of an image. Semantic segmentation is used to identify sets of pixels that constitute distinguishable categories. When applied to defect identification of photovoltaic modules, semantic segmentation involves segmenting the defect shape (pixels with defect labels) from the original image for further analysis.

[0076] Intersection over Union (IoU): This calculates the ratio of the intersection to the union of two bounding boxes. Generally, if IoU ≥ 0.5, the detection result is acceptable and the detection is correct. If IoU equals 1, since the intersection equals the union, it indicates that the predictor perfectly overlaps with the actual bounding box. A threshold of 0.5 is typically set to determine if the predicted bounding box is correct; the higher the IoU, the more accurate the bounding box and the more accurate the recognition.

[0077] like Figure 2 As shown in the figure, a photovoltaic module defect detection method provided by an embodiment of the present invention includes the following steps:

[0078] Step 1: Determine the optimal thermal excitation parameters and the optimal lift-off distance.

[0079] To ensure the safety and reliability of the thermal excitation parameters, this invention determines the optimal thermal excitation parameters before applying thermal excitation to the photovoltaic module using a thermal excitation source. In this embodiment, the thermal excitation source is an induction coil, and the thermal excitation parameters include the excitation current and the thermal excitation application time. When an induction coil is used as the thermal excitation source, the thermal excitation application time is the same as the electromagnetic induction excitation application time. In a specific embodiment of this invention, the specific implementation process for determining the optimal thermal excitation parameters and the optimal lift-off distance includes:

[0080] Step 1.1: Construct a coupled simulation model consisting of an induction coil model and a photovoltaic module model.

[0081] Step 1.2: Using the coupled simulation model, obtain the first relationship curve between the temperature of the upper surface of the photovoltaic module model and the electromagnetic induction excitation application time under different excitation currents, and the second relationship curve between the eddy current density on the surface of the cell inside the photovoltaic module model and the lift-off distance; where the lift-off distance h is the vertical distance between the induction coil model and the upper surface of the photovoltaic module model.

[0082] Step 1.3: The excitation current magnitude and thermal excitation application time corresponding to the first relationship curve that takes the shortest time to heat the surface temperature of the photovoltaic module model from the first temperature to the second temperature are the optimal thermal excitation parameters; in the second relationship curve, the lift-off distance corresponding to the maximum eddy current density is the optimal lift-off distance.

[0083] In step 1.1, the finite element analysis method is used, and a coupled simulation model consisting of an induction coil model and a photovoltaic module model is constructed using finite element simulation software such as COMSOL and ANSYS. Figure 3 As shown, the induction coil model is located directly above the photovoltaic module model. a is the width of the single-layer rectangular induction coil, b is the length of the single-layer rectangular induction coil, and h is the lift-off distance.

[0084] In step 1.2, based on the coupled simulation model, the lift-off distance is set and kept constant. The magnitude of the excitation current of the induction coil model is changed, and different electromagnetic induction excitations are applied to the photovoltaic module model. The first relationship curve of the temperature on the upper surface of the photovoltaic module model under different excitation currents as a function of the electromagnetic induction excitation application time is obtained.

[0085] Each excitation current corresponds to a first relationship curve, and each first relationship curve is a curve showing the change in temperature of the upper surface of the photovoltaic module model as a function of the electromagnetic induction excitation application time. When the temperature of the upper surface of the photovoltaic module model reaches a set temperature value, the electromagnetic induction excitation is stopped. The set temperature value is less than the maximum value of the photovoltaic module's operating temperature range; in this embodiment, the set temperature value is 65°C.

[0086] In this embodiment, the lift-off distance is set to 5cm when the first relationship curve is obtained.

[0087] Based on the coupled simulation model, the excitation parameters of the induction coil model are set according to the optimal thermal excitation parameters in step 1.3 and kept constant. By changing the lift-off distance, the second relationship curve of the eddy current density on the surface of the cell inside the photovoltaic module model as a function of the lift-off distance is obtained.

[0088] In step 1.3, both the first and second temperatures are within the operating temperature range of the photovoltaic module, and the second temperature is lower than the set temperature value. Assuming the operating temperature range of the photovoltaic module is -40℃ to 85℃, and since the subsequent test platform is built at room temperature, the first temperature in this embodiment is set to 25℃, and the second temperature is set as far away from the maximum value of the operating temperature range as possible; therefore, the second temperature in this embodiment is set to 60℃. That is, among all the first relationship curves, the first relationship curve that requires the shortest time to heat the surface temperature of the photovoltaic module model from 25℃ to 60℃ corresponds to the optimal thermal excitation parameters in terms of excitation current magnitude and electromagnetic induction excitation application time. The first relationship curve corresponding to the optimal thermal excitation parameters is the optimal relationship curve.

[0089] In the second relationship curve, the lift-off distance corresponding to the maximum eddy density is the optimal lift-off distance.

[0090] To verify the effectiveness of the coupled simulation model and ensure the accuracy of the optimal thermal excitation parameters and optimal lift-off distance determined by the coupled simulation model, between steps 1.2 and 1.3, the method of the present invention further includes a step of verifying the first relationship curve and the second relationship curve. The specific verification process is as follows:

[0091] a: Construct the first experimental platform consisting of physical induction coils, thermal image acquisition equipment, and physical photovoltaic modules;

[0092] b: Based on the first experimental platform, under the same excitation current, electromagnetic induction excitation application time and lift-off distance as the obtained first relationship curve, electromagnetic induction excitation is applied to the photovoltaic module, and the first experimental curve of the temperature of the upper surface of the photovoltaic module under different excitation currents as electromagnetic induction excitation application time is obtained; that is, each excitation current corresponds to a first experimental curve, and each first experimental curve is a curve of the temperature of the upper surface of the photovoltaic module model as electromagnetic induction excitation application time.

[0093] c: Based on the first test platform, under the same excitation current, electromagnetic induction excitation application time (i.e., optimal thermal excitation parameters) and lift-off distance as the second relationship curve obtained, electromagnetic induction excitation is applied to the photovoltaic module to obtain the second test curve of the eddy current density on the surface of the cell inside the photovoltaic module as a function of lift-off distance.

[0094] d: Compare the first relationship curve with the first experimental curve. If the trends of the first relationship curve and the first experimental curve are consistent, and the relative error between the first relationship curve and the first experimental curve is less than the error threshold, then the first relationship curve is verified. Otherwise, proceed to step 1.2 and re-simulate to obtain the first relationship curve.

[0095] Compare the second relationship curve with the second experimental curve. If the trends of the second relationship curve and the second experimental curve are consistent, and the relative error between the second relationship curve and the second experimental curve is less than the error threshold, then the second relationship curve is verified. Otherwise, proceed to step 1.2 and re-simulate to obtain the second relationship curve.

[0096] In this embodiment, the formula for calculating the relative error is:

[0097]

[0098] Where r is the relative error, x is the temperature value on the first relationship curve or the eddy current density value on the second relationship curve, and x' is the temperature value on the first test curve or the eddy current density value on the second test curve. In this embodiment, the error threshold is 10% to 15%.

[0099] Step 2: Determine the image extraction time period based on the optimal thermal excitation parameters and the optimal lift-off distance.

[0100] To reduce the workload of image selection from the image sequence and improve defect detection efficiency in step 4, the image extraction time period is determined based on the optimal thermal excitation parameters and optimal lift-off distance determined in step 1. The specific implementation process is as follows:

[0101] Step 2.1: Construct a second experimental platform consisting of an induction coil, an external light source, photovoltaic modules, and a light image acquisition device; wherein the photovoltaic modules are defective photovoltaic modules;

[0102] Step 2.2: Set the excitation current and electromagnetic induction excitation application time of the induction coil in the second test platform according to the optimal thermal excitation parameters determined in Step 1.3, and determine the distance between the induction coil and the photovoltaic module according to the optimal lift-off distance;

[0103] Step 2.3: Simultaneously apply electromagnetic induction excitation and optical excitation to the photovoltaic module using an induction coil and an external light source, while the optical image acquisition device simultaneously acquires the optical image sequence of the photovoltaic module;

[0104] Step 2.4: Evaluate the sharpness of each frame in the light image sequence to obtain the third relationship curve between the average gray value of the image and the thermal excitation application time;

[0105] Step 2.5: If the optimal relationship curve is positively correlated with the third relationship curve, then select the later thermal excitation application time period in the optimal relationship curve as the image extraction time period;

[0106] If the optimal relationship curve is negatively correlated with the third relationship curve, then the thermal excitation application time period earlier in the optimal relationship curve is selected as the image extraction time period.

[0107] If the optimal relationship curve and the third relationship curve are partially positively correlated and partially negatively correlated, then the time period of thermal excitation application corresponding to the positive correlation in the optimal relationship curve is selected as the image extraction time period.

[0108] Traditional photovoltaic power plant inspection methods often employ dual-light imaging with thermal infrared and visible light cameras. This approach can only detect issues like component damage and shading that produce localized hot spots, failing to identify fault modes such as microcracks and PID effects that do not cause temperature changes. In this embodiment, a short-wave infrared camera is used instead of a visible light camera to detect light radiation generated by photoluminescence, thus increasing the range of photovoltaic module defects that can be detected using traditional inspection methods.

[0109] In step 2.4, the Tenengrad gradient function is used to evaluate the sharpness of each frame in the light image sequence. The specific implementation process is as follows:

[0110] Step 2.41: Use the Sobel operator to extract the horizontal and vertical gradients for each pixel in each frame of the light image sequence;

[0111] Step 2.42: Calculate the sum of squares of the gradients in the horizontal and vertical directions for each pixel to obtain the gradient value for each pixel;

[0112] Step 2.43: Calculate the average grayscale value of each frame of the image based on the gradient value of each pixel;

[0113] Step 2.44: Calculate the average gray value of each frame in the light image sequence to obtain the third relationship curve. The specific calculation formula is as follows:

[0114]

[0115] Where Ten is the average grayscale value of each frame, G(x,y) is the gradient value at pixel (x,y), and n is the total number of pixels in the image. G(x,y) is defined as:

[0116]

[0117] Among them, G x (x,y) represents the gradient value in the horizontal direction at pixel (x,y), G y (x, y) represents the gradient value in the vertical direction at pixel (x, y), defined as follows:

[0118]

[0119]

[0120] in, Let f(x,y) be the convolution symbol, f(x,y) be the gray value at pixel (x,y), and g be the gray value at pixel (x,y). x g y These are the horizontal and vertical operators of the Sobel operator, respectively, as follows:

[0121]

[0122]

[0123] Short-wave infrared images are a type of grayscale image. Grayscale refers to the color depth of a pixel in a black and white image, typically ranging from 0 to 255, with white being 255 and black being 0. Therefore, black and white images are also called grayscale images.

[0124] In step 2.5, when the optimal relationship curve is positively correlated with the third relationship curve, it indicates that the higher the temperature of the upper surface of the photovoltaic module, the higher the clarity of the light image, and the later the light image is, the clearer it is. Therefore, the later thermal excitation application time period in the optimal relationship curve is selected as the image extraction time period in step 4.

[0125] For example, assuming the optimal thermal excitation application time determined by the optimal relationship curve is the time period from 0 to T, when the thermal excitation source is set according to the optimal thermal excitation parameters and the optimal lift-off distance, thermal and optical excitation are applied to the photovoltaic module simultaneously using both the thermal and optical excitation sources. After simultaneously acquiring the thermal and optical image sequences of the photovoltaic module, the thermal and optical image sequences corresponding to the acquisition time and image extraction time period are selected for filtering, reducing the filtering workload while ensuring the clarity of the thermal and optical images. In this embodiment, the later thermal excitation application time period is from T / 3 to T, and the thermal and optical image sequences from the time period from 0 to T / 3 are discarded.

[0126] When the optimal relationship curve is negatively correlated with the third relationship curve, it indicates that the higher the temperature of the photovoltaic module's surface, the lower the clarity of the light image, and the clearer the light image is further forward. Therefore, the earlier thermal excitation application time period in the optimal relationship curve is selected as the image extraction time period in step 4. Let the optimal thermal excitation application time determined by the optimal relationship curve be the time period from 0 to T. In this embodiment, the earlier thermal excitation application time period is the time period from 0 to 2T / 3, and the thermal image sequence and light image sequence of the time period from 2T / 3 to T are discarded.

[0127] When the optimal relationship curve and the third relationship curve are partially positively correlated and partially negatively correlated (including first positively correlated and then negatively correlated, or first negatively correlated and then positively correlated), the time period of thermal excitation application corresponding to the positive correlation is selected as the image extraction time period.

[0128] For example, suppose the optimal thermal excitation application time determined by the optimal relationship curve is the time period from 0 to T. The optimal relationship curve for the time period from 0 to T / 3 is positively correlated with the third relationship curve, and the optimal relationship curve for the time period from T / 3 to T is negatively correlated with the third relationship curve. Then, the time period from 0 to T / 3 is selected as the image extraction time period.

[0129] Step 3: Set up the thermal excitation source according to the optimal thermal excitation parameters and optimal lift-off distance determined in Step 1. Apply thermal excitation and optical excitation to the photovoltaic module simultaneously using the thermal excitation source and the optical excitation source, and simultaneously acquire the thermal image sequence and optical image sequence of the photovoltaic module.

[0130] To obtain an image containing defects, the photovoltaic module is a photovoltaic module with defects.

[0131] In this embodiment, a thermal image acquisition device is used to acquire a sequence of thermal images of the photovoltaic module, and an optical image acquisition device is used to acquire a sequence of optical images of the photovoltaic module. The thermal image acquisition device is a thermal imager, and the optical image acquisition device is a short-wave infrared camera. The thermal excitation source is an induction coil, and the optical excitation source is an external light source. The application time of the optical excitation source is the same as the application time of the thermal excitation source, and the optical excitation source is perpendicular to the photovoltaic module. While applying thermal and optical excitation to the photovoltaic module, the thermal and optical image sequences of the photovoltaic module are acquired simultaneously, ensuring synchronization between the excitation application time and the image sequence acquisition time.

[0132] Electromagnetic induction is used to heat photovoltaic (PV) modules, inducing currents (eddy currents) in the cells inside. According to Joule's law, some of this induced current is converted into heat, which is then conducted into the surrounding space as thermal radiation, creating a three-dimensional heat flow field and affecting the surface temperature of the PV module. Simultaneously, photoluminescence technology applies an external light source to the PV module, influencing the carrier balance within the cells and generating light radiation. Photoluminescence is a technique that uses external light sources to illuminate PV modules, causing carrier transitions and producing light radiation. It allows for simultaneous imaging and detection of surface and internal defects in the PV modules.

[0133] This invention replaces the traditional photovoltaic (PV) module inspection solution that typically combines a thermal imager with a visible light camera with a short-wave infrared camera. The short-wave infrared camera detects the light radiation generated by photoluminescence from the PV modules, thus increasing the variety of defects detected in traditional inspection methods. Furthermore, the thermal image sequences acquired by the thermal imager exhibit thermal diffusion, which may affect the accuracy of electromagnetic induction thermal imaging (EMI) defect detection. Photoluminescence detection technology can also assist EPI thermal imaging in feature extraction. By fusing and extracting features from the multimodal images acquired by the thermal imager and the short-wave infrared camera, the accuracy of defect detection can be improved.

[0134] Traditional photovoltaic (PV) module thermal imaging inspection methods are static inspections that rely on the quality of transient thermal images of the PV module at a specific moment. This invention considers the dynamic temperature changes during the electromagnetic induction heating process of the PV module, establishing a time-series optical and thermal image set of the heating process. This avoids the impact of poor quality in a single transient image on the detection results. Furthermore, different defects have varying effects on the photothermal radiation changes of the PV module; monitoring the time-series optical and thermal images of the PV module allows for more accurate defect detection.

[0135] Step 4: Extract the thermal image sequence and optical image sequence corresponding to the acquisition time and the image extraction time period determined in Step 2 from the thermal image sequence and optical image sequence obtained in Step 3. Filter the extracted thermal image sequence and optical image sequence respectively to obtain a clear thermal image set and optical image set corresponding to the acquisition time.

[0136] If the application time of the thermal excitation source and the optical excitation source is 0 to T, and the image extraction time period is 0 to T / 3, then the thermal image sequence and the optical image sequence with the acquisition time of 0 to T / 3 can be extracted from the thermal image sequence and the optical image sequence obtained in step 3.

[0137] Thermal and optical image sequences may contain distorted images, such as tilted images or unclear defect features. These distorted images need to be removed during image screening to obtain clear thermal and optical images. During feature fusion, thermal and optical images acquired at the same time need to be fused. Therefore, among the clear thermal and optical images, images acquired at the same time that only contain thermal or optical images need to be removed, ensuring that both thermal and optical images exist simultaneously at the same acquisition time.

[0138] Step 5: Perform feature fusion on the thermal image and optical image acquired at the same time to obtain a photothermal feature fusion image, and then obtain a photothermal feature fusion image set.

[0139] Optical images in a set of optical images do not exhibit lateral thermal diffusion and can clearly display defect outlines; however, defect features such as microcracks are not obvious in optical images. Thermal images in a set of thermal images accurately reflect defect locations, but determining the specific type of defect is more difficult. Feature fusion of thermal and optical images combines the advantages of both, improving the accuracy of defect type and location detection.

[0140] In this embodiment, feature fusion of thermal and optical images acquired at the same time is performed using a deep convolutional generative adversarial network (GAN)-based method. The specific implementation process includes:

[0141] Step 5.1: Align the thermal and optical images acquired at the same time to achieve registration between the thermal and optical images;

[0142] Step 5.2: Simultaneously input the registered thermal image and optical image acquired at the same time into the generator network G to obtain a single fused image. Then, simultaneously input the single fused image and the label image into the discriminator network D. Through adversarial training, complete the binary classification task to obtain a high-quality optical-thermal feature fused image.

[0143] In this network, the generator network G connects the optical and thermal images using a concatenate function. High-dimensional features are extracted from the images via residual unit blocks to obtain a single fused image. The label image is generated by dividing the registered thermal and optical images acquired at the same time into sub-blocks with a fixed stride using a sliding window. A fusion rule is then applied, selecting the higher gradient from the high-frequency components and weighting the low-frequency components, followed by a Laplacian pyramid transform. The discriminator network D uses stride convolutions to extract depth features from both the single fused image and the label image, and employs the LeakyReLU activation function to enhance the network's nonlinearity. Finally, a fully connected layer outputs the optical-thermal feature fused image.

[0144] Step 6: Label each image in the photothermal feature fusion image set from Step 5 with the corresponding attribute labels.

[0145] In this embodiment, the attribute labels include defect type and defect location, and all labeled images constitute a sample dataset. The sample dataset is divided into a training set and a validation set in a 7:3 ratio to train the defect detection model.

[0146] Step 7: Train the defect detection model using images with attribute labels to obtain the target defect detection model.

[0147] The defect detection model uses existing models, such as neural network models and YOLOv5 models. During training, images from the training set are used as input to the defect detection model, and the attribute labels corresponding to those images are used as the target output. Semantic segmentation is performed on defects in the images to achieve quantification and classification of defects. After training, a preliminary defect detection model is obtained. This preliminary model can segment potentially defective shapes from the original image in the photothermal feature fusion image, and then compare them with defect shapes in the labeled image to identify the possible types of defects and their confidence levels.

[0148] The detection capability of the initially trained defect detection model on new samples is tested using a validation set. When the intersection-union ratio (IUU) of the defect types detected by the defect detection model and the manually labeled attribute tags is greater than 0.5, the defect detection model is considered to have completed training, and the target defect detection model is obtained.

[0149] Step 8: When the image of the photovoltaic module to be inspected is acquired, the defect detection result corresponding to the image of the photovoltaic module to be inspected is determined using the target defect detection model.

[0150] Similar to the samples used during training, the image of the photovoltaic module to be detected is obtained by fusing the features of the thermal image and the optical image of the photovoltaic module.

[0151] This invention uses an electromagnetic induction coil to electromagnetically excite photovoltaic (PV) modules during nighttime periods when they are not generating electricity, generating induced currents (eddy currents) inside the PV modules. Due to the resistive effect of the PV modules, the eddy currents generate Joule heat inside the PV modules, actively heating them. Therefore, it is possible to detect the temperature change process of the PV module surface temperature from ambient temperature to a specified temperature, achieving non-steady-state thermal imaging detection. Field of view and spatial resolution are two important parameters of thermal imagers. The field of view refers to the maximum spatial range that a thermal imager can detect. A larger field of view allows the thermal imager to capture a wider area, but the actual size of the object corresponding to each pixel also increases, leading to a grainier image (reduced spatial resolution). The field of view is proportional to the product of spatial resolution and pixels. Since the number of pixels in a thermal imager is a fixed value, the closer the thermal imager is to the object being detected, the smaller the field of view, and the higher the corresponding spatial resolution.

[0152] During temperature changes in photovoltaic (PV) modules, the difference in thermal conductivity between areas with early defects and small-area defects such as cracks and scratches and normal areas affects the internal heat transfer process of the PV module, thus influencing the temperature distribution on the module surface. Therefore, by combining the temperature distribution change process on the PV module surface with photoluminescence detection technology and using deep learning algorithms to analyze the defect characteristics of abnormal temperature distribution areas, it can be determined whether the module has early or small-area defects.

[0153] This invention, based on the law of electromagnetic induction, uses an induction coil to heat a photovoltaic (PV) module, generating corresponding photothermal time-series images. Defect features are extracted from these images, and the impact of defects on the PV module's photothermal radiation is analyzed, thus achieving intelligent defect detection in PV modules. This invention simultaneously applies light and heat excitation to the PV module. Under these excitations, the module's temperature rises, generating a continuous sequence of light and heat radiation images. The light and heat radiation images are fused, and then defect detection is performed on the fused image. Combining the advantages of thermal imaging and short-wave infrared imaging, this invention expands the types of defects that can be detected in PV modules and improves the accuracy of defect detection. It can detect early-stage defects and small-area defects, overcoming the limitations of traditional thermal imaging detection techniques, such as difficulty in detecting small-area defects and the limited range of image modes. This invention offers advantages such as non-contact operation, detailed evaluation, rapid detection, and intelligent detection, providing a reliable and rapid detection method for the efficient operation and maintenance of PV module arrays in large-scale PV power plants, and contributing to enhancing the innovation capabilities of my country's PV power generation industry.

[0154] like Figure 4 and 5 As shown, this embodiment of the invention also provides a photovoltaic module defect detection system, including:

[0155] An induction coil 3 and an external light source 2 are positioned above the photovoltaic module 5. The induction coil 3 and the external light source 2 are used to simultaneously apply electromagnetic induction excitation and light excitation to the photovoltaic module 5.

[0156] Thermal imager 4 is used to acquire a sequence of thermal images of photovoltaic module 5 when electromagnetic induction excitation is applied to photovoltaic module 5;

[0157] A short-wave infrared camera 1 is used to acquire a sequence of light images of the photovoltaic module 5 when light excitation is applied to the photovoltaic module 5.

[0158] A data processing device is connected to the thermal imager 4 and the shortwave infrared camera 1, respectively. The data processing device is used to acquire thermal image sequences and optical image sequences of the photovoltaic module 5; to filter the thermal image sequences and optical image sequences to obtain clear thermal image sets and optical image sets corresponding to the acquisition time; to perform feature fusion on the thermal images and optical images at the same acquisition time to obtain photothermal feature fusion images, and then obtain a photothermal feature fusion image set; to label each image in the photothermal feature fusion image set with corresponding attribute labels; to train the defect detection model using the images with attribute labels to obtain the target defect detection model; when acquiring the image of the photovoltaic module to be detected, the target defect detection model is used to determine the defect detection result corresponding to the image of the photovoltaic module to be detected.

[0159] The solar cell 53 is housed within the encapsulating film 54 and tempered glass 52. An external light source 2 and an induction coil 3 apply optical and electromagnetic induction excitation to the photovoltaic module 5, respectively. The induction coil 3 generates a changing induced current 55 (eddy current) in the photovoltaic module 5 in a non-contact manner through electromagnetic induction excitation. The presence of a defect 56 affects the eddy current distribution, eddy current heating, and heat conduction processes, leading to differences in the temperature field distribution on the surface of the photovoltaic module 5. The external light source 2 continuously applies optical excitation to the photovoltaic module 5, causing the charge carriers in the solar cell 53 inside the photovoltaic module 5 to absorb energy and generate light radiation 51. Simultaneously with the application of electromagnetic induction excitation and optical excitation, an infrared thermal imager 4 and a short-wave infrared camera 1 begin acquiring a thermal infrared time-series image sequence (i.e., thermal image sequence) and a short-wave infrared light time-series image sequence (i.e., light image sequence) generated by the photovoltaic module 5. The thermal imager 4 records the change in the surface temperature field of the photovoltaic module 5 with the application of excitation, forming a thermal infrared time-series image sequence of the photovoltaic module 5's temperature rise process. The short-wave infrared camera 1 acquires short-wave infrared time-series images of the photovoltaic module 5.

[0160] The excitation parameters and lift-off distance of the induction coil 3 are set according to the optimal thermal excitation parameters and optimal lift-off distance determined in step 1.3 of the photovoltaic module defect detection method as described above. The application time of the external light source 2 is the same as the application time of the electromagnetic induction excitation generated by the induction coil 3.

[0161] Before filtering the thermal image sequence and the optical image sequence, the thermal image sequence and the optical image sequence corresponding to the image extraction time period determined in step 2 of the photovoltaic module defect detection method described above are extracted from the acquired thermal image sequence and optical image sequence. Then, the extracted thermal image sequence and optical image sequence are filtered to obtain a clear thermal image set and an optical image set corresponding to the acquisition time.

[0162] In one specific embodiment of the present invention, the induction coil 3, the external light source 2, the thermal imager 4, and the short-wave infrared camera 1 are mounted on a robotic arm, which is mounted on a car. The car equipped with the robotic arm is used for convenient movement of the induction coil 3, the external light source 2, the thermal imager 4, and the short-wave infrared camera 1, for power supply, and for setting the lift-off distance and excitation parameters of the induction coil 3 according to the optimal thermal excitation parameters and the optimal lift-off distance.

[0163] The above description only discloses specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or modifications that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting defects in photovoltaic modules, characterized in that, The detection method includes the following steps: Simultaneously apply thermal and optical excitation to the photovoltaic module, and simultaneously acquire thermal and optical image sequences of the photovoltaic module; The thermal image sequence and the optical image sequence are filtered to obtain a clear set of thermal images and a set of optical images that correspond to the acquisition time. By fusing features of thermal and optical images acquired at the same time, a set of optical-thermal feature fusion images is obtained. Each image in the photothermal feature fusion image set is labeled with a corresponding attribute tag; The defect detection model is trained using images with attribute labels to obtain the target defect detection model; When an image of a photovoltaic module to be inspected is acquired, the target defect detection model is used to determine the defect detection result corresponding to the image of the photovoltaic module to be inspected. The detection method further includes, before applying thermal and light excitation to the photovoltaic module simultaneously: determining the optimal thermal excitation parameters and the optimal lift-off distance, and setting the thermal excitation source according to the optimal thermal excitation parameters and the optimal lift-off distance to apply thermal excitation to the photovoltaic module. The specific implementation process for determining the optimal thermal excitation parameters and the optimal lift-off distance includes: A coupled simulation model consisting of a thermal excitation source model and a photovoltaic module model is constructed, with the thermal excitation source model located directly above the photovoltaic module model; Using the coupled simulation model, a first relationship curve between the temperature of the upper surface of the photovoltaic module model and the thermal excitation application time under different excitation currents is obtained, as well as a second relationship curve between the eddy current density on the surface of the cell inside the photovoltaic module model and the lift-off distance; wherein, the lift-off distance refers to the vertical distance between the thermal excitation source model and the upper surface of the photovoltaic module model; The optimal thermal excitation parameters are the excitation current magnitude and thermal excitation application time corresponding to the first relationship curve that minimizes the time required to heat the upper surface of the photovoltaic module model from the first temperature to the second temperature; wherein, both the first temperature and the second temperature are within the operating temperature range of the photovoltaic module; and in the second relationship curve, the lift-off distance corresponding to the maximum eddy current density is the optimal lift-off distance.

2. The photovoltaic module defect detection method according to claim 1, characterized in that, The first temperature is 25°C, and the second temperature is 60°C.

3. The photovoltaic module defect detection method according to claim 1, characterized in that, The specific process for obtaining the first relationship curve is as follows: By setting a lift-off distance and keeping it constant, changing the magnitude of the excitation current of the thermal excitation source model, and applying different thermal excitations to the photovoltaic module model, a first relationship curve is obtained showing the temperature of the upper surface of the photovoltaic module model changing with the thermal excitation application time under different excitation currents. The specific process for obtaining the second relationship curve is as follows: Set the thermal excitation parameters of the thermal excitation source model and keep the thermal excitation parameters unchanged, change the lift-off distance, and obtain the second relationship curve of the eddy current density on the surface of the cell inside the photovoltaic module model as a function of the lift-off distance; The thermal excitation parameters are the optimal thermal excitation parameters.

4. The photovoltaic module defect detection method according to claim 1, characterized in that, Before determining the optimal thermal excitation parameters based on the first relationship curve and the optimal lift-off distance based on the second relationship curve, the detection method further includes a step of verifying the first and second relationship curves, the specific implementation process of which is as follows: The first experimental platform was constructed, consisting of a physical thermal excitation source, thermal image acquisition equipment, and a physical photovoltaic module. Based on the first test platform, under the same thermal excitation parameters and lift-off distance as the first relationship curve, thermal excitation was applied to the photovoltaic module to obtain a first test curve showing the change of the temperature of the upper surface of the photovoltaic module with the thermal excitation application time under different excitation currents. Based on the first test platform, under the same thermal excitation parameters and lift-off distance as the obtained second relationship curve, thermal excitation was applied to the photovoltaic module to obtain a second test curve showing the change of eddy current density on the surface of the cell inside the photovoltaic module with the lift-off distance. The first relationship curve is compared with the first experimental curve. If the relative error between the first relationship curve and the first experimental curve is less than the error threshold, the first relationship curve is verified. Otherwise, the simulation is repeated to obtain the first relationship curve. The second relationship curve is compared with the second experimental curve. If the relative error between the second relationship curve and the second experimental curve is less than the error threshold, the second relationship curve is verified. Otherwise, the simulation is repeated to obtain the second relationship curve.

5. The photovoltaic module defect detection method according to claim 1, characterized in that, Before simultaneously applying thermal and optical excitation to the photovoltaic module, the detection method further includes: determining the image extraction time period based on the optimal thermal excitation parameters and the optimal lift-off distance, the specific implementation process of which includes: A second experimental platform was constructed, consisting of a thermal excitation source, an external light source, photovoltaic modules, and optical image acquisition equipment; wherein the photovoltaic modules were defective. The thermal excitation source of the second test platform is set according to the optimal thermal excitation parameters and the optimal lift-off distance; The photovoltaic module is simultaneously subjected to thermal excitation and light excitation using the thermal excitation source and an external light source, and the light image sequence of the photovoltaic module is simultaneously acquired using a light image acquisition device. The sharpness of each frame in the optical image sequence is evaluated to obtain a third relationship curve between the average gray value of the image and the thermal excitation application time. If the optimal relationship curve is positively correlated with the third relationship curve, then the later thermal excitation application time period in the optimal relationship curve is selected as the image extraction time period. If the optimal relationship curve is negatively correlated with the third relationship curve, then the earlier thermal excitation application time period in the optimal relationship curve is selected as the image extraction time period. If the optimal relationship curve and the third relationship curve are partially positively correlated and partially negatively correlated, then the thermal excitation application time period corresponding to the positive correlation in the optimal relationship curve is selected as the image extraction time period. Wherein, the optimal relationship curve is the first relationship curve corresponding to the optimal thermal excitation parameters determined by the coupled simulation model; Before filtering the thermal image sequence and the optical image sequence, thermal image sequence and optical image sequence corresponding to the acquisition time and the image extraction time period are extracted from the thermal image sequence and the optical image sequence for filtering.

6. The photovoltaic module defect detection method according to claim 5, characterized in that, The Tenengrad gradient function is used to evaluate the sharpness of each frame in the optical image sequence. The specific implementation process is as follows: The Sobel operator is used to extract the horizontal and vertical gradients for each pixel in each frame of the light image sequence; The gradient value of each pixel is obtained by summing the squares of the horizontal and vertical gradients. The average grayscale value of each frame of the image is calculated based on the gradient value of each pixel. The average gray value of each frame in the light image sequence is calculated to obtain the third relationship curve.

7. The photovoltaic module defect detection method according to claim 1, characterized in that, The feature fusion of thermal and optical images acquired at the same time using a deep convolutional generative adversarial network (GAN) method is specifically implemented as follows: Alignment operations are performed on thermal and optical images acquired at the same time to achieve registration between thermal and optical images; The registered thermal image and optical image acquired at the same time are input into the generator network G to obtain a single fused image. Then, the single fused image and the label image are input into the discriminator network D to obtain a photothermal feature fused image. In this network, the generator network G connects the optical and thermal images using a concatenate function, extracts high-dimensional features from the images through residual unit blocks, and obtains a single fused image. The label image is generated by dividing the registered thermal and optical images acquired at the same time into sub-blocks using a sliding window, and then using a fusion rule that selects the larger gradient of the high-frequency part and performs a weighted average of the low-frequency part, and using Laplacian pyramid transformation. The discriminator network D uses stride-length convolution to extract the depth features of the single fused image and the label image respectively, and uses the LeakyReLU activation function to enhance the nonlinearity of the network. Finally, the optical and thermal feature fused image is output as a fully connected layer.

8. A photovoltaic module defect detection system, characterized in that, The system includes: A thermal excitation source and a light excitation source are disposed above the photovoltaic module, and the thermal excitation source and the light excitation source are used to simultaneously apply thermal excitation and light excitation to the photovoltaic module; Thermal image acquisition equipment is used to acquire a sequence of thermal images of photovoltaic modules when thermal excitation is applied to them; An optical image acquisition device is used to acquire a sequence of optical images of a photovoltaic module when optical excitation is applied to it. A data processing device is connected to the thermal image acquisition device and the optical image acquisition device, respectively. The data processing device is used to simultaneously acquire thermal image sequences and optical image sequences of the photovoltaic module; to filter the thermal image sequences and optical image sequences to obtain clear thermal image sets and optical image sets corresponding to the acquisition time; to perform feature fusion on the thermal images and optical images acquired at the same time to obtain a photothermal feature fusion image, and further to obtain a photothermal feature fusion image set; to label each image in the photothermal feature fusion image set with corresponding attribute tags; to train a defect detection model using the images with attribute tags to obtain a target defect detection model; and when acquiring an image of the photovoltaic module to be detected, to use the target defect detection model to determine the defect detection result corresponding to the image of the photovoltaic module to be detected. Before applying thermal and light excitation to the photovoltaic module simultaneously, the data processing device is also used to determine the optimal thermal excitation parameters and the optimal lift-off distance, so as to set the thermal excitation source according to the optimal thermal excitation parameters and the optimal lift-off distance to apply thermal excitation to the photovoltaic module. The specific implementation process for determining the optimal thermal excitation parameters and the optimal lift-off distance includes: A coupled simulation model consisting of a thermal excitation source model and a photovoltaic module model is constructed, with the thermal excitation source model located directly above the photovoltaic module model; Using the coupled simulation model, a first relationship curve between the temperature of the upper surface of the photovoltaic module model and the thermal excitation application time under different excitation currents is obtained, as well as a second relationship curve between the eddy current density on the surface of the cell inside the photovoltaic module model and the lift-off distance; wherein, the lift-off distance refers to the vertical distance between the thermal excitation source model and the upper surface of the photovoltaic module model; The optimal thermal excitation parameters are the excitation current magnitude and thermal excitation application time corresponding to the first relationship curve that minimizes the time required to heat the upper surface of the photovoltaic module model from the first temperature to the second temperature; wherein, both the first temperature and the second temperature are within the operating temperature range of the photovoltaic module; and in the second relationship curve, the lift-off distance corresponding to the maximum eddy current density is the optimal lift-off distance.

Images