Photovoltaic module defect detection system and method based on multi-physical quantity fusion

Abstract

The invention relates to a photovoltaic module defect detection system and method based on multi-physical quantity fusion. The method comprises the following steps: acquiring dust layer optical data and dust weighing data of the photovoltaic module; calculating a cleanliness compensation coefficient of the photovoltaic module according to the ash layer optical data and the dust weighing data; according to the cleanliness compensation coefficient, performing data correction on the original electroluminescent image of the photovoltaic module to obtain a corrected electroluminescent image; inputting the corrected electroluminescent image into a defect cascade neural model of the photovoltaic module to obtain image defect analysis data; and according to the image defect analysis data and the physical constraint conditions, performing defect delimitation on the photovoltaic module to obtain a defect detection result of the photovoltaic module. By adopting the method, the requirements of efficient and accurate detection can be met.

Description

Technical Field

[0001] The present application relates to the technical field of photovoltaic module detection, and in particular to a photovoltaic module defect detection system and method based on multi-physical quantity fusion. Background Art

[0002] In traditional technologies, photovoltaic module inspection mainly relies on a single detection technology, such as electroluminescence (EL) detection and infrared thermal imaging. However, these technologies have problems such as low detection accuracy, poor real-time performance, and insufficient adaptability to complex environments. Especially in large-scale photovoltaic power stations, it is difficult to meet the needs of efficient and accurate detection. Summary of the Invention

[0003] Based on this, it is necessary to provide a photovoltaic module defect detection system and method based on multi-physical quantity fusion that can meet the needs of efficient and accurate detection in response to the above technical problems.

[0004] In a first aspect, the present application provides a photovoltaic module defect detection method based on multi-physical quantity fusion, comprising:

[0005] Obtain optical data of dust layer and dust weighing data of photovoltaic modules;

[0006] Calculating a cleanliness compensation coefficient of the photovoltaic module according to the ash layer optical data and the dust weighing data;

[0007] performing data correction on the original electroluminescent image of the photovoltaic module according to the cleanliness compensation coefficient to obtain a corrected electroluminescent image;

[0008] inputting the corrected electroluminescent image into a defect cascade neural model of the photovoltaic module to obtain image defect analysis data;

[0009] Defects of the photovoltaic module are delineated according to the image defect analysis data and physical constraints to obtain defect detection results of the photovoltaic module.

[0010] In a second aspect, the present application also provides a photovoltaic module defect detection system based on multi-physical quantity fusion, the system comprising a computer device and a terminal,

[0011] The terminal is used to obtain optical data of the dust layer and dust weighing data of the photovoltaic module and transmit them to the computer device;

[0012] The computer device is used to calculate the cleanliness compensation coefficient of the photovoltaic module based on the dust layer optical data and the dust weighing data;

[0013] The computer device is used to perform data correction on the original electroluminescent image of the photovoltaic module according to the cleanliness compensation coefficient to obtain a corrected electroluminescent image;

[0014] The computer device is configured to input the corrected electroluminescent image into a defect cascade neural model of the photovoltaic module to obtain image defect analysis data;

[0015] The computer device is used to perform defect delineation on the photovoltaic module based on the image defect analysis data and physical constraints to obtain a defect detection result of the photovoltaic module.

[0016] In a third aspect, the present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements any step of the photovoltaic module defect detection method based on multi-physical quantity fusion.

[0017] The above-mentioned photovoltaic module defect detection system and method based on multi-physical quantity fusion obtains the optical transmittance of the photovoltaic module's dust layer and dust weighing data to construct a refined cleanliness compensation coefficient and perform dynamic brightness correction on the original electroluminescent image, effectively eliminating the impact of dust obstruction on image quality and improving the visualization of micro-defects such as cracks and broken grids. Furthermore, a cascaded neural network model with the ability to jointly identify hidden cracks and broken grids is adopted to achieve collaborative detection and feature extraction of multiple types of defects, avoiding misjudgments and information loss caused by a single model. Furthermore, a physical constraint mechanism based on thermal field-optical field consistency judgment is combined to perform conflict verification and logical correction on the detection results, ensuring the accuracy and physical rationality of the output defect judgment results. This system not only meets the needs of efficient and accurate detection in complex environments, but also maintains a high recognition rate and outputs comprehensive diagnostic results including defect type, distribution, severity, and repair priority, with enhanced environmental robustness, system reliability, and operation and maintenance guidance value. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a diagram illustrating an application environment of a photovoltaic module defect detection method based on multi-physical quantity fusion in one embodiment;

[0020] Figure 21 is a flow chart of a photovoltaic module defect detection method based on multi-physical quantity fusion in one embodiment;

[0021] Figure 3 Schematic diagram of a flow chart of a method for calculating a cleanliness compensation coefficient in one embodiment;

[0022] Figure 4 A schematic flow chart of a method for constructing a three-dimensional dust density distribution model in one embodiment;

[0023] Figure 5 Schematic diagram of a flow chart of a method for constructing transmittance distribution data in one embodiment;

[0024] Figure 6 1 is a flow chart of a method for calculating mass density information in one embodiment;

[0025] Figure 7 1 is a flow chart of a method for obtaining image defect analysis data in one embodiment;

[0026] Figure 8 1 is a flow chart of a method for obtaining component broken gate detection data in one embodiment;

[0027] Figure 9 1 is a flow chart of a method for obtaining component broken gate detection data in another embodiment;

[0028] Figure 10 Schematic diagram of a flow chart of a method for obtaining defect detection results in one embodiment;

[0029] Figure 11 A system architecture diagram of a photovoltaic module defect detection method based on multi-physical quantity fusion in one embodiment;

[0030] Figure 12 1. A schematic diagram of the entire process of a photovoltaic module defect detection method based on multi-physical quantity fusion in one embodiment;

[0031] Figure 13 1 is a schematic diagram of an implementation flow of a photovoltaic module defect detection method based on multi-physical quantity fusion in one embodiment;

[0032] Figure 14 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0034] The photovoltaic module defect detection method based on multi-physical quantity fusion provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store data that server 104 needs to process. The data storage system can be integrated with server 104 or placed on a cloud or other network server. Server 104 can be implemented as a standalone server or a server cluster consisting of multiple servers.

[0035] In an exemplary embodiment, Figure 2 As shown in the figure, a photovoltaic module defect detection method based on multi-physical quantity fusion is provided. Figure 1 The server in FIG. 1 is taken as an example to illustrate the method, including the following steps 202 to 210. Among them:

[0036] Step 202: Obtain optical data of the dust layer and dust weighing data of the photovoltaic module.

[0037] Step 204 : Calculate the cleanliness compensation coefficient of the photovoltaic module based on the dust layer optical data and the dust weighing data.

[0038] Step 206 : performing data correction on the original electroluminescent image of the photovoltaic module according to the cleanliness compensation coefficient to obtain a corrected electroluminescent image.

[0039] In step 208 , the corrected electroluminescent image is input into a defect cascade neural model of the photovoltaic module to obtain image defect analysis data.

[0040] Step 210 : Delineate defects of the photovoltaic module based on the image defect analysis data and physical constraints to obtain defect detection results of the photovoltaic module.

[0041] A photovoltaic module is an energy conversion device consisting of multiple photovoltaic cells connected in series or parallel. It converts solar energy into direct current (DC) electricity and is a core component of a photovoltaic power generation system. Its structure, comprising layers such as the glass panel, encapsulation material, solar cells, and backsheet, is susceptible to factors such as dust accumulation, mechanical damage, and ribbon breakage, which can lead to reduced output performance.

[0042] Among them, the dust layer optical data can be the transmittance or reflectance information obtained by irradiating the surface of the photovoltaic module with laser or visible light, which is used to reflect the degree of blocking of the optical signal by the dust layer. The data is usually expressed in the form of a pixel matrix.

[0043] The dust weighing data may be data obtained by measuring the dust mass on the surface of the photovoltaic module through a micro-force sensor array or a high-precision weighing module, and is used to quantitatively reflect the physical load of dust per unit area.

[0044] The cleanliness compensation coefficient is an image correction parameter calculated based on dust optical data and dust weighing data. It represents the degree of brightness attenuation at each image pixel due to dust obstruction. This coefficient is used to compensate electroluminescent images or other images, improving image quality and defect recognition accuracy.

[0045] Raw electroluminescence images are generated by injecting a certain current into a photovoltaic module in the absence of external light, revealing hidden defects such as cracks in the module's material and broken solder joints. These images are not compensated or enhanced in any way and may be affected by factors such as dust layers and uneven lighting.

[0046] Among them, the corrected electroluminescent image can be an image generated by applying image enhancement algorithms such as the cleanliness compensation coefficient to perform reverse brightness correction on the original EL image. It has higher contrast and texture clarity and can more effectively expose structural defects such as tiny cracks and broken grids.

[0047] Among them, the defect cascade neural model can be a deep learning detection framework that integrates a multi-stage neural network structure, which is usually composed of a hidden crack recognition sub-model and a broken grid recognition sub-model, which respectively handle different types of image feature extraction tasks.

[0048] Among them, the image defect analysis data can be the structured recognition results output by the defect cascade neural model, including characteristic indicators such as crack area, broken grid position, defect confidence, crack length, and broken grid number.

[0049] Physical constraints can be judgment rules based on actual physical mechanisms introduced during defect analysis, such as that hot spots should match cracks or broken gate areas, and that strong EL signals should not exist in gray layer areas. These conditions are used to verify the consistency of multimodal inspection results, eliminate potential artifacts or conflicting results, and enhance the physical rationality of judgments.

[0050] Among them, the defect detection results can be based on the fusion of image defect analysis data and physical constraints, clearly indicating the specific defect type, location, severity and priority in the photovoltaic module. The results are used to guide subsequent module repair, replacement or monitoring strategies.

[0051] Specifically, the surface of the photovoltaic module is synchronously collected through a laser transmittance measurement system and a micro-force sensor array: a laser array with a wavelength of 532nm is used to scan the module surface, and a high-sensitivity photodiode is used to obtain the transmitted light intensity of each pixel area, thereby constructing a transmittance image as the optical data of the gray layer; at the same time, a vibration mechanism is used to make the dust on the surface of the module fall off and fall into the weighing tray, and the total weight of the dust is measured by the micro-force sensor as the dust weighing data.

[0052] Based on the optical data of the dust layer, the transmittance reduction rate of each area is calculated, reflecting the degree of optical obstruction of the local dust layer. Based on the measured dust weight data, combined with the sensor sensing area and the transmittance attenuation ratio, the dust weight is spatially weighted to estimate the mass density per unit area of ​​each area. Furthermore, by integrating the above optical and mass data, a three-dimensional dust density distribution model is constructed. The mapping relationship between dust thickness and electroluminescent image (EL) brightness is used to solve the three-dimensional dust density distribution model and calculate the cleanliness compensation coefficient corresponding to each pixel.

[0053] During image processing, the cleanliness compensation coefficient matrix is ​​first mapped to the original EL image coordinate system to achieve pixel-level registration. Subsequently, brightness inverse compensation is performed on each pixel value in the original EL image to enhance the contrast of the original EL image based on the principle that "the thicker the dust layer, the greater the signal attenuation." This process can be pre-processed by a dynamic enhancement module (such as the Dust Compensation Layer) before the neural network input, effectively improving the discernibility of faint defect features and generating a corrected EL image with significantly improved visual quality.

[0054] The compensated electroluminescence image is fed into a cascaded neural network model consisting of two submodules: a crack detection subnetwork based on the DeepLabv3+ architecture, which segments and identifies small cracks and estimates their length and area; and a fault detection subnetwork, based on an improved version of YOLOv8, which uses the Hough transform to locate fault lines and accurately detect the position and length of faults within a specified window. The model's final output includes image defect analysis data, including the location distribution of cracks and faults, characteristic indicators, and confidence levels.

[0055] Combining image defect analysis results with a physical constraint model under module operating conditions, multi-source consistency verification and composite defect decision-making are performed. If a crack is detected without a corresponding hot spot temperature rise or an artifact signal is present in an image-occluded area, a physical conflict verification mechanism is triggered to prevent misjudgment. If a crack is detected, along with a corresponding hot spot temperature rise or the absence of an artifact signal in an image-occluded area, a physical conflict verification mechanism is triggered to prevent misjudgment. A defect scoring model is constructed based on parameters such as crack length, number of broken gates, and hot spot temperature difference, comprehensively outputting the severity, spatial distribution, and repair priority of each defect. Ultimately, defect detection results for the PV module are generated, including defect type, impact level, and maintenance recommendations, for system operation and maintenance.

[0056] In one embodiment, Figure 11 This is a system architecture diagram of a photovoltaic module defect detection method based on multi-physical quantity fusion; Figure 12 A schematic diagram of the full process integration of the photovoltaic module defect detection method based on multi-physical quantity fusion; Figure 13 Schematic diagram of the implementation process of the photovoltaic module defect detection method based on the fusion of multiple physical quantities.

[0057] In the above-mentioned photovoltaic module defect detection method based on multi-physical quantity fusion, a refined cleanliness compensation coefficient is constructed by obtaining the optical transmittance of the photovoltaic module's dust layer and dust weighing data. Dynamic brightness correction is then performed on the original electroluminescent image, effectively eliminating the impact of dust obstruction on image quality and improving the visualization of micro-defects such as cracks and broken grids. Furthermore, a cascaded neural network model with the ability to jointly identify hidden cracks and broken grids is adopted to achieve collaborative detection and feature extraction of multiple types of defects, avoiding misjudgments and information loss caused by a single model. Furthermore, a physical constraint mechanism based on thermal field-optical field consistency judgment is combined to perform conflict verification and logical correction on the detection results, ensuring the accuracy and physical rationality of the output defect judgment results. This method not only meets the requirements for efficient and accurate detection in complex environments, but also maintains a high recognition rate and outputs comprehensive diagnostic results including defect type, distribution, severity, and repair priority, demonstrating enhanced environmental robustness, system reliability, and operational guidance value.

[0058] In an exemplary embodiment, Figure 3 As shown, the calculation of the cleanliness compensation coefficient of the photovoltaic module based on the dust layer optical data and the dust weighing data includes steps 302 to 304.

[0059] Step 302: construct a three-dimensional dust density distribution model based on the dust layer optical data and dust weighing data.

[0060] Step 304 : solving the three-dimensional dust density distribution model based on the light scattering physical model of the photovoltaic module to obtain the cleanliness compensation coefficient of the photovoltaic module.

[0061] The three-dimensional dust density distribution model can be constructed by integrating optical transmittance data from the PV module surface with dust weighing data to create a mathematical model that reflects the spatial distribution characteristics of dust on the module surface. This model not only describes the dust mass density per unit area in each region but also, by inferring the three-dimensional accumulation pattern of dust on the module surface based on the spatial variation trend of the dust layer thickness.

[0062] The light scattering physical model is a mathematical description system based on the interaction between light and dust particles, used to simulate the scattering and attenuation effects of dust on the propagation of light signals in electroluminescent images. This model typically combines Mie scattering theory or empirical fitting functions, taking into account factors such as particle size, density, refractive index, and thickness, to establish a functional relationship between dust density and image brightness loss.

[0063] Specifically, laser transmittance measurements are used to obtain transmittance drop images from different regions of the photovoltaic module surface, reflecting the degree of light obstruction by dust. Simultaneously, a micro-force sensor array is used to measure the total mass of dust removed from the module surface. Based on the degree of transmittance drop in each region, a weighted distribution algorithm is used to distribute the total dust mass to corresponding regions, forming a mass density map per unit area. Combining the sensor's sensing area geometry with known dust density parameters, a thickness inversion is performed for each sampling region. Ultimately, a three-dimensional density distribution model of dust on the photovoltaic module surface is generated, characterizing the spatial volume distribution of dust.

[0064] After obtaining a three-dimensional dust density distribution model, a light scattering physical model was established based on the principles of optical scattering and absorption. This model, which considers the effects of dust particle size, density, and thickness on the intensity attenuation of the EL image, was used to simulate or perform regression analysis on the optical propagation of the EL image. This model established a nonlinear mapping relationship between dust density and image brightness attenuation. Based on this mapping function, the corresponding brightness attenuation value was calculated for each image pixel, forming a cleanliness compensation coefficient matrix for subsequent pixel-level compensation of the EL image.

[0065] In this embodiment, a three-dimensional dust density distribution model is constructed based on dust layer optical data and dust weighing data. This distribution model is then solved in conjunction with a light scattering physics model. This allows for the precise quantification of the spatial accumulation of dust on the surface of photovoltaic modules and its impact on the optical attenuation of electroluminescent images, thereby generating a physically meaningful cleanliness compensation coefficient. This compensation coefficient corrects brightness distortion caused by dust occlusion on a pixel-by-pixel basis, effectively improving the visual contrast of defective areas and the accuracy of neural network recognition. This significantly enhances the system's adaptability to complex environments such as dusty and highly reflective environments, overcoming the technical limitations of traditional single-image enhancement methods, which suffer from poor robustness to environmental changes.

[0066] In an exemplary embodiment, Figure 4 As shown, the construction of the dust density three-dimensional distribution model based on the dust layer optical data and the dust weighing data includes steps 402 to 406.

[0067] Step 402 : constructing transmittance distribution data corresponding to the surface of the photovoltaic module based on the gray layer optical data.

[0068] Step 404 : Calculate the mass density information of the dust on the surface of the photovoltaic module based on the dust weighing data and the sensor sensing area of ​​the photovoltaic module.

[0069] Step 406 : Fusing the transmittance distribution data and the mass density information to obtain a three-dimensional dust density distribution model.

[0070] The transmittance distribution data may be a two-dimensional image matrix formed by the ratio of the transmitted light intensity at different positions to the reference value obtained after optically scanning the surface of the photovoltaic module, which is used to reflect the degree to which light in each area is blocked by the dust layer.

[0071] The sensor sensing area may be the physical surface area range covered or responded to by each micro-force sensor in the dust weighing device, which is usually pre-determined based on the sensor installation position and component structure.

[0072] Mass density information can be the mass of dust per unit area, reflecting the spatial distribution of dust on the surface of the PV module. It is calculated based on the total dust weight obtained by the micro-force sensor and weighted by the sensor's sensing area to generate a two-dimensional mass density map.

[0073] Specifically, the server controls a laser scanner to illuminate the PV panel surface point by point, recording the transmitted light intensity at each pixel. This intensity is then compared to a pre-set dust-free reference image at the pixel level to determine the relative transmittance of each region. This transmittance data is then smoothed and spatially interpolated to create a two-dimensional transmittance distribution map covering the entire PV panel surface. This map serves as the transmittance distribution data, reflecting the degree of optical signal obstruction caused by dust in each region.

[0074] The total dust mass data obtained by the micro-force sensor array is spatially mapped to the sensing area corresponding to each sensor. The total weight is weighted and distributed to each area according to the dust level or predefined area ratio in each area, and the dust mass density per unit area is calculated. This density information constitutes a preliminary two-dimensional mass distribution map as mass density information.

[0075] By spatially aligning the transmittance distribution map with the mass density map and establishing an inverse functional relationship between dust thickness and transmittance attenuation, the conversion from dust mass density to volume density (thickness) is achieved. At each pixel location, the vertical thickness of the dust accumulation is calculated based on the mass density value and the inverse functional relationship model, ultimately forming a three-dimensional dust density distribution model with spatial coordinates, thickness values, and optical properties.

[0076] In this embodiment, by constructing transmittance distribution data for the photovoltaic module surface based on optical data from the dust layer, combining dust weighing data with dust mass density information calculated from the sensor sensing area, and fusing the two to generate a three-dimensional dust density distribution model, the system can accurately reconstruct the dust coverage state on the photovoltaic module surface. This not only overcomes the resolution and accuracy limitations of single optical or weighing methods, but also effectively identifies localized non-uniformities and accumulation trends in dust distribution. The resulting three-dimensional distribution model provides critical data support for subsequent image compensation, defect development, and environmental adaptability enhancement, significantly improving the robustness and detection reliability of the photovoltaic module defect detection system under complex operating conditions.

[0077] In an exemplary embodiment, Figure 5 As shown, the process of constructing the transmittance distribution data corresponding to the surface of the photovoltaic module based on the gray layer optical data includes steps 502 to 506.

[0078] Step 502 : Control the laser array corresponding to the photovoltaic module to scan the surface of the photovoltaic module to obtain the transmitted light intensity distribution data of the photovoltaic module.

[0079] Step 504 : Perform a pixel-level ratio operation on the transmitted light intensity distribution data and the dust-free reference image of the photovoltaic module to obtain a transmittance matrix.

[0080] Step 506 : Perform filtering and interpolation processing on the transmittance matrix to generate transmittance distribution data.

[0081] The laser array may be a scanning device composed of a plurality of controllable laser emitting units arranged in a regular pattern, used to emit a directional laser beam to the surface of a photovoltaic module to achieve rapid optical scanning of a large area.

[0082] Transmitted light intensity distribution data can be a collection of transmitted light intensity values ​​at various locations, as light passes through the dust layer on the surface of a photovoltaic module under laser illumination and is received and recorded by the back-mounted photoelectric sensor. This data, presented as a two-dimensional matrix or image, reflects the degree of light obstruction in different areas and serves as an important input for calculating the transmittance matrix and analyzing dust distribution.

[0083] The dust-free reference image may be a standard transmission image obtained by the same laser scanning method when the surface of the photovoltaic module is completely clean, and is used as a comparison benchmark with the actual detection image.

[0084] Among them, the transmittance matrix can be a two-dimensional data structure generated by performing a pixel-by-pixel ratio operation on the transmitted light intensity distribution data obtained by the photovoltaic component in the actual detection state and its reference image in the dust-free state, wherein each matrix element represents the relative transmittance value of the corresponding position.

[0085] Specifically, a laser array with a preset wavelength (e.g., 532nm) is controlled to scan the surface of a photovoltaic module in an orderly manner. The laser beam illuminates the module surface point by point and penetrates the overlying dust layer. A photosensor array located on the back of the module simultaneously records the light intensity value transmitted at each scanning point. The system collects the responses of all laser points using a time-synchronized mechanism, generating a two-dimensional distribution of the transmitted light intensity at every location on the photovoltaic module surface as transmitted light intensity distribution data, which reflects the degree of light flux obstruction by the dust layer.

[0086] The transmitted light intensity distribution data is compared with the pre-stored dust-free transmission reference image of the same component collected in a clean state, and the pixel-by-pixel ratio is calculated. That is, the transmittance value of each pixel position is calculated, which indicates the degree to which the light intensity at that position is maintained in the current state compared to the clean state. The transmittance matrix is ​​formed, which directly quantifies the differences and relative severity of dust obstruction in spatial distribution.

[0087] To address possible noise points, sensor errors, or spatial sampling discontinuities in the transmittance matrix, image filtering algorithms (such as Gaussian filtering or mean filtering) are applied to smooth the transmittance matrix and remove local abnormal data. At the same time, bilinear interpolation or spline interpolation methods are used to fill in low-density sampling areas to generate a complete and continuous high-precision transmittance distribution map as the transmittance distribution data.

[0088] In this embodiment, a laser array is controlled to precisely scan the surface of a photovoltaic module to obtain transmitted light intensity distribution data. This data is then compared to a dust-free reference image and compared to a pixel-by-pixel ratio to generate a transmittance matrix. Combined with filtering and interpolation, this method yields highly accurate transmittance distribution data. This not only enables detailed spatial characterization of dust obstruction, but also significantly improves the integrity and noise immunity of the detected data. Compared to traditional single-point photometry or image grayscale estimation techniques, this method offers the advantages of higher resolution, more accurate local identification, and greater environmental adaptability, providing more reliable optical data support for subsequent dust density modeling and image compensation.

[0089] In an exemplary embodiment, Figure 6 As shown, the calculation of the mass density information of the dust on the surface of the photovoltaic module based on the dust weighing data and the sensor sensing area of ​​the photovoltaic module includes steps 602 to 606.

[0090] Step 602 : obtaining a total dust mass value of the photovoltaic module based on the dust mass of the photovoltaic module sensed by the sensor sensing area.

[0091] Step 604 : dividing the data sampling areas according to the spatial mapping relationship of the sensor sensing areas on the surface of the photovoltaic module.

[0092] Step 606 : Allocate the total dust mass value to the surface of the photovoltaic module according to the change rate of the transmittance distribution data in each data sampling area to obtain mass density information.

[0093] Among them, dust mass perception can be a process of real-time monitoring of the mass of dust shed due to factors such as vibration, wind or cleaning through micro-force sensors arranged on the surface of or under the photovoltaic module.

[0094] The total dust mass value may be the overall mass value of the dust currently accumulated on the surface of the photovoltaic module obtained by adding up the dust weight signals collected by all sensors.

[0095] Among them, the spatial mapping relationship can be a one-to-one correspondence model established between each sensor sensing area and the actual geometric position of the photovoltaic module surface, which is used to correctly associate the collected local physical data with the two-dimensional coordinate system of the component.

[0096] The data sampling area may be a plurality of sub-areas divided on the surface of the photovoltaic module according to the sensor layout and its corresponding spatial mapping relationship, and each area corresponds to a sensor or a group of pixel ranges.

[0097] Specifically, an array of micro-force sensors is deployed on the surface of a photovoltaic module or in a collection tray area beneath it. When vibration or natural disturbances cause dust to fall off the module surface, the sensors collect real-time weight change data. The local mass values ​​measured by each sensor are combined to determine the total mass of dust on the PV module surface, providing a total mass benchmark for subsequent regional distribution modeling.

[0098] Based on the layout structure of the sensors and the geometric dimensions of the photovoltaic modules, a one-to-one mapping relationship is established between the sensing area corresponding to each sensor and the position on the module surface. Based on this relationship, the module surface is divided into multiple independent data sampling areas, so that each area can be traced back to a specific sensor, realizing the spatial attribution of weight data and laying a spatial resolution foundation for the calculation of local mass density.

[0099] The total dust mass is weighted and distributed, combining the rate of change in the transmittance distribution within each data sampling area (transmittance distribution data) as a reference indicator of dust thickness or accumulation. Regions with more significant decreases in dust layer transmittance are assigned higher mass weights, resulting in more dust mass being allocated to those regions. Finally, the allocated mass for each region is divided by its corresponding area to calculate the dust mass density, creating a two-dimensional density map reflecting the degree of surface contamination.

[0100] In this embodiment, dust mass is sensed on the surface of photovoltaic modules based on the sensor's sensing area, obtaining the total dust mass. The spatial mapping of the sensing area on the module surface is then used to divide the data sampling area into refined areas. The total mass is then weighted based on the rate of change of the transmittance distribution data in each area, enabling quantitative reconstruction of the dust distribution on the module surface. This method transcends the limitations of traditional dust detection, which can only provide global averages or single-point judgments, and can generate spatially resolved mass density maps that effectively reflect local differences in contamination and accumulation trends. This provides more accurate data support for image compensation, defect identification, and cleaning and maintenance strategies, significantly improving the accuracy and practicality of the detection system.

[0101] In an exemplary embodiment, Figure 7 As shown, the step of inputting the corrected electroluminescent image into the defect cascade neural model of the photovoltaic module to obtain image defect analysis data includes steps 702 to 706.

[0102] Step 702 : Perform component hidden crack defect detection on the corrected electroluminescent image to obtain component hidden crack detection data.

[0103] Step 704 : Perform component broken gate defect detection on the corrected electroluminescent image to obtain component broken gate detection data.

[0104] Step 706 : Coordinately process the component hidden crack detection data and the component broken grid detection data to obtain image defect analysis data.

[0105] Among them, component hidden crack defect detection can be a process of identifying fine cracks in photovoltaic cells caused by stress, thermal expansion and contraction or manufacturing defects by performing image enhancement and deep learning recognition processing on electroluminescent images.

[0106] Among them, the component hidden crack detection data can be the structured recognition results output by the hidden crack detection model, which usually includes information such as the location coordinates, length, direction, coverage area, grayscale contrast and the severity level derived from the crack.

[0107] Among them, component broken gate defect detection can refer to the process of identifying local conductive path interruptions caused by mechanical damage or manufacturing anomalies through edge extraction and breakpoint recognition algorithms for the conductive gate line structure in the electroluminescent image.

[0108] Among them, the component broken gate detection data can be a multi-parameter information set output by the broken gate detection model, including the broken gate position coordinates, length, gate line number, whether it is located in a high current density area, confidence score, etc.

[0109] Among them, collaborative processing can be the process of jointly analyzing the hidden crack detection data and the broken grid detection data. Based on the physical correlation between defects (such as the co-occurrence of cracks and broken grids) and indicators such as spatial overlap, a defect impact superposition model or voting mechanism is established to output more comprehensive and more reliable image defect analysis results.

[0110] Specifically, the cleanliness-compensated corrected electroluminescence image is fed into a deep convolutional neural network model for module subtle crack defect detection. Using architectures such as DeepLabv3+, the model extracts crack features and identifies tiny subtle cracks within the cell. This model is trained to learn the crack's edge morphology, grayscale gradients, and spatial structure. It can accurately locate crack areas after significantly reducing grayscale interference, and outputs subtle crack detection data including crack location, length, direction, and crack impact level.

[0111] In the modified electroluminescent image, the Canny edge detection algorithm is first used to extract the gate line boundaries. The Hough transform is then applied to identify the position and direction of each gate line, delineating candidate gate line windows. Within these windows, an improved YOLOv8 model is then used to identify breakpoints and extract parameters such as the number, location, and length of broken gates. Combined with the gate line's current path properties and location information, structured component broken gate detection data is generated, serving as a crucial basis for assessing electrical performance risks.

[0112] The component hidden crack detection data and the component broken grid detection data are input into the defect fusion analysis module, and the spatial overlap between the crack and the broken grid, the electrical impact weight and the composite defect severity analysis are calculated through the preset physical coupling model. At the same time, the consistency verification logic is executed to eliminate image artifacts and conflicting annotation information, and finally the image defect analysis data including defect type, location distribution, confidence level and repair suggestions are output.

[0113] In this embodiment, by performing component hidden crack defect detection and component broken grid defect detection on the corrected electroluminescent image separately, and collaboratively processing the two types of detection data, it is not only possible to accurately identify key defects such as tiny cracks and conductive grid line breaks, but also to determine the severity of complex defects through spatial coupling and physical logical relationships between defects, effectively improving the comprehensiveness and accuracy of defect identification. This breaks through the limitations of traditional single-model detection, which is prone to missed detection and misjudgment, and significantly enhances the system's comprehensive analysis capabilities in handling complex defect scenarios (such as cracks accompanied by broken grids and hot spot occlusion), providing image defect analysis results with more diagnostic value for subsequent operation and maintenance decisions.

[0114] In an exemplary embodiment, Figure 8 As shown, the process of performing component broken gate defect detection on the corrected electroluminescent image to obtain component broken gate detection data includes steps 802 to 808.

[0115] Step 802: Perform Kenny edge detection on the corrected electroluminescent image to obtain device gate line boundary information.

[0116] Step 804 : determining the position and direction of the element gate line boundary information according to the Hough transform algorithm, and obtaining candidate windows of each gate line region.

[0117] Step 806 : Identify the breakpoint position of each gate line region candidate window, and extract each gate break length and the corresponding gate break position.

[0118] Step 808 : Match each gate-break length and the corresponding gate-break position with the current density distribution model of the component to determine the component gate-break detection data.

[0119] Among them, Kenny edge detection can be an image processing algorithm used to detect edge areas with drastic brightness changes in images, with high positioning accuracy and noise resistance.

[0120] The element gate line boundary information may be a collection of data on the position, direction and edge shape of the gate line contour extracted from the electroluminescent image by an image edge detection algorithm, and is usually expressed in the form of pixel coordinates and line segment descriptions.

[0121] The Hough transform algorithm is a mathematical method used to detect geometric shapes such as lines and circles in images. It is particularly suitable for extracting image features with clear geometric structures. In photovoltaic module grid line detection, the Hough transform accurately identifies the direction, length, and position of grid lines by mapping point sets in edge images into parameter space.

[0122] The boundary information can be the edge feature data between the object and the background in the image. In this system, it specifically refers to the bright and dark boundary information extracted from the corrected EL image, which is used to reflect the position contours of structures such as grid lines and cracks.

[0123] The candidate window for the gate line region may be a rectangular region of a certain width and length constructed around each gate line in the image based on the gate line position information obtained by Hough transform, and is used to limit the calculation range of broken gate detection.

[0124] Specifically, the Canny edge detection algorithm is applied to the cleanliness-compensated corrected electroluminescent image to extract areas with distinct bright and dark boundaries, focusing on identifying the edge contours of the gate lines. The Canny algorithm calculates image gradients, performs non-maximum suppression, and performs double-threshold processing. This algorithm can preserve the continuity of the gate line edges while suppressing noise interference, thereby generating a clear gate line boundary map as the device's gate line boundary information.

[0125] Based on the device's gate line boundary information, the probabilistic Hough line transform algorithm is used to identify linear gate line structures in the image and calculate their start and end coordinates, length, and orientation angle. By establishing positioning information for each gate line, a candidate gate line region window centered around it is generated. This is used to limit the image area range for broken gate detection, resulting in candidate gate line regions.

[0126] Within each candidate window of a gate line region, an improved YOLOv8 object detection network optimized for gate break defects is applied to identify and locate breakpoints in the image. The model is trained to learn the texture fracture characteristics and brightness mutation patterns of gate breaks, accurately outputting the break's location coordinates and pixel-level length information. Each gate break is recorded with its start and end locations, fracture extent, and confidence level, forming a structured gate break identification result.

[0127] The coordinates of the broken gate identification results are aligned with the component current density distribution model to determine whether each broken gate is located on a high-current transmission path, and the importance is judged in combination with the fracture length threshold. If the broken gate length exceeds the set value and falls in the critical current area, it is marked as a high-risk broken gate. Based on the broken gate identification results, combined with the structural layout of the photovoltaic module and the current density distribution model, the position and length of each broken gate are spatially matched and analyzed to determine whether it is located in a high-current path area or a key node of the electrode. The degree of its impact on the conductive performance is calculated based on factors such as the broken gate length, position sensitivity, and local current intensity, thereby achieving a quantitative analysis of the physical impact of the broken gate and generating component broken gate detection data.

[0128] In this embodiment, Kenny edge detection is performed on the corrected electroluminescent image to extract grid line boundary information, Hough transform is used to accurately locate the grid line direction and generate candidate grid line area windows, and then a deep recognition algorithm is used to finely identify the breakpoints within each window. The extracted grid break position and length are matched and analyzed with the current density distribution model. This can achieve a full process evaluation from structural image features to electrical performance impact. This not only improves the positioning accuracy and recognition robustness of grid break detection, but also effectively identifies high-risk grid breaks on critical paths of electrical functions, significantly enhancing the system's ability to identify major hidden dangers and its warning value. Compared with traditional image morphology-based judgment methods, it has greater engineering practicality and diagnostic depth.

[0129] In an exemplary embodiment, Figure 9 As shown, the step of matching each of the broken gate lengths and the corresponding broken gate positions with the current density distribution model of the component to determine the broken gate detection data of the component includes steps 902 to 906.

[0130] Step 902 : Calculate a two-dimensional current density distribution diagram of the photovoltaic module based on the designed operating parameters and actual operating status information of the photovoltaic module.

[0131] Step 904 : For any gate-breaking position, calculate the angle between the gate line direction of the gate-breaking length and the main current direction to obtain the gate-breaking direction angle.

[0132] Step 906 : Calculate the component gate-break detection data according to the gate-break direction angle, the gate-break length, and the local current density.

[0133] Among them, the design operating parameters may refer to the structure and electrical settings of the photovoltaic module under ideal working conditions, including the number and arrangement of solar cells, series-parallel topology, standard operating voltage and current, wiring method, electrode material and grid line layout, etc.

[0134] Actual operating status information can be the actual operating conditions of PV panels at a specific point in time, typically including ambient temperature, light intensity (irradiance), panel voltage and current output, shading conditions, and operating time. This information is acquired through on-site data collection equipment (such as data loggers and temperature / light sensors) and is used to modify theoretical model parameters and generate dynamic current distribution diagrams that conform to real-world operating conditions.

[0135] The two-dimensional current density distribution map can be a spatial distribution image formed by discretely dividing the surface of the photovoltaic module into several sub-regions and solving the current intensity per unit area of ​​each region based on structural parameters and operating status.

[0136] The broken gate angle can be the angle between the extension direction of the broken gate identified in the component in the image and the main current direction in the area. This angle quantifies the degree to which the broken gate blocks the current transmission path: the closer the angle is to 90 degrees, the more significant the interference of the broken gate on the current path.

[0137] The local current density refers to the current per unit area at a specific location or small area on the surface of a photovoltaic module, usually expressed in A / cm 2 Its value is determined by many factors, including component structure, charge collection path, environmental radiation, and fault impact. It is a key parameter for evaluating the electrical load intensity of a certain area and determining the severity of the impact of a gate failure.

[0138] Specifically, a current density model is developed based on the design and operating parameters of the PV module (such as the cell layout, electrode connection method, and rated operating parameters (such as voltage, current, and inter-cell interconnection method)), combined with real-time operating status data (such as irradiance, temperature, and current output). This model uses the finite element method or meshing algorithm to divide the module surface into multiple sub-regions. Taking into account the series-parallel topology, resistance distribution, and charge-carrying characteristics, it outputs a two-dimensional current density map reflecting the distribution of current intensity across the module surface. This map is used to determine the conductive load sensitivity at different locations.

[0139] For any of the aforementioned broken gate locations, after identifying the location and extension direction of the broken gate, the strike vector of the gate line where the broken gate is located is extracted. The mainstream current direction vector at that location is fitted from the two-dimensional current density map. The angle between the broken gate direction and the main current direction is calculated using the angle calculation formula between the two vectors. This angle reflects the degree to which the broken gate interrupts the current conduction path and is an important directional indicator for subsequent electrical impact assessments.

[0140] By comprehensively considering the length and orientation angle of the fault, as well as the local current density in the area where it is located, an influence function or weighting model is used to quantify the impact of the fault on the overall current transmission efficiency. For example, if the fault is nearly perpendicular to the main current flow direction and located in a high current density area, it will have a greater impact on the conductivity. This calculation process can determine the risk level and impact weight of each fault, ultimately generating structural component fault detection data that includes information such as location, size, and electrical impact level.

[0141] In this embodiment, by combining the design and operating parameters of photovoltaic modules with actual operating conditions, a two-dimensional current density distribution map is calculated. The angle between the fault direction and the main current direction is further analyzed, and the impact of the fault length and local current density are comprehensively quantified. This allows for a precise assessment of each fault defect in terms of structural location, electrical function, and conduction risk. This method breaks through the limitations of traditional fault detection based solely on image morphology and introduces physical electrical parameters as a criterion. This significantly improves the engineering credibility and electrical performance correlation of fault detection results, providing a more valuable foundation for defect screening, fault prioritization, and the development of refined operation and maintenance strategies.

[0142] In an exemplary embodiment, Figure 10 As shown, the process of performing defect delineation on the photovoltaic module based on the image defect analysis data and the physical constraint conditions to obtain the defect detection result of the photovoltaic module includes steps 1002 to 1006.

[0143] Step 1002 : verifying conflict information between the image defect analysis data and each sensor data according to physical constraints.

[0144] Step 1004 , when the conflict information is not a null value, return to the step of obtaining the optical data of the dust layer of the photovoltaic module and the dust weighing data, until the conflict information is a null value.

[0145] Step 1006 : Based on the image defect analysis data, confidence voting is performed on each preset defect of the photovoltaic module to obtain a defect detection result.

[0146] Among them, sensor data can refer to real-time or historical information collected by various physical sensors installed in photovoltaic modules or their operating environment, including parameters such as current, voltage, temperature, thermal imaging, light intensity, vibration, etc.

[0147] The conflicting information may be the logical inconsistency between the image defect analysis results and the physical sensor data, such as a broken gate identified in the image but no current attenuation, or a hot spot appearing in the infrared image while there is no crack in the image.

[0148] Pre-defect confidence assessment refers to the process of assigning credibility scores to various standardized defect types defined in the system (such as hidden cracks, broken grids, hot spots, and contamination) based on multiple sources, including image recognition results, physical data verification, and model prediction probabilities. Each defect type is scored based on multi-dimensional evidence, and its existence and impact level are comprehensively judged using a voting mechanism, confidence fusion, or weighted average method.

[0149] Specifically, after completing image defect analysis, the system uses operating status data collected by external physical sensors (such as current, voltage, temperature, or infrared thermal imaging) to perform a data consistency comparison with the image defect analysis data based on preset physical constraints (such as thermal anomalies corresponding to cracks, current decay corresponding to broken gates, etc.). If the comparison data indicates a logical contradiction between the detected defect location or nature and the physical behavior reflected by the sensor, it is marked as conflict information and used to evaluate the reliability of the detection results.

[0150] If the comparison data indicates an inconsistency between the image defect analysis data and the physical sensor data—that is, if the conflict information is not null—the backtracking mechanism is automatically triggered to reacquire the dust layer optical transmittance data and dust weighing data to update the cleanliness compensation coefficient and re-correct the electroluminescent image. This process repeats the compensation and image re-recognition process until the conflict information is eliminated or the consistency condition is met, ensuring that the final inspection result is based on a high degree of consistency between the optical, physical, and environmental data.

[0151] After all defect types are identified, confidence-weighted statistics are calculated for each pre-defined defect based on its source (e.g., EL image model, infrared thermal image, broken gate identification, etc.). A voting mechanism is then used to integrate the detection results from multiple data sources. Ultimately, the validity and severity of each defect are determined based on the number of votes, confidence threshold, and data consistency. A comprehensive defect detection result is then output, including location, type, and confidence level.

[0152] In this embodiment, physical constraints are introduced to perform consistency checks on image defect analysis data and multi-source sensor data, and the image compensation process is automatically backtracked and corrected when conflicting information exists, ensuring that the image recognition results are highly consistent with the actual physical state, thereby significantly improving the credibility and engineering applicability of defect recognition. Furthermore, by implementing a confidence voting mechanism that integrates multimodal data for various preset defects, image recognition, physical verification, and model output information are effectively integrated, enabling defect recognition to transition from single-image judgment to multi-source intelligent decision-making. This method not only enhances the system's adaptability to complex operating conditions and edge cases, but also significantly reduces the probability of false alarms and missed detections, providing more reliable data support for intelligent diagnosis and refined operation and maintenance of photovoltaic modules.

[0153] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0154] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 14 As shown. The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. Those skilled in the art will understand that Figure 14 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0155] In one embodiment, a photovoltaic module defect detection system based on multi-physical quantity fusion is also provided, the system including a computer device and a terminal.

[0156] A terminal is used to obtain optical data of the dust layer and dust weighing data of the photovoltaic module and transmit them to a computer device;

[0157] Computer equipment, used to calculate the cleanliness compensation coefficient of the photovoltaic module based on the optical data of the dust layer and the dust weighing data;

[0158] A computer device, used for performing data correction on an original electroluminescent image of the photovoltaic module according to a cleanliness compensation coefficient to obtain a corrected electroluminescent image;

[0159] a computer device for inputting the corrected electroluminescent image into a defect cascade neural model of the photovoltaic module to obtain image defect analysis data;

[0160] Computer equipment is used to delineate defects of photovoltaic modules based on image defect analysis data and physical constraints to obtain defect detection results of photovoltaic modules.

[0161] In one embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0162] In one embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the steps in the above-mentioned method embodiments when executed by a processor.

[0163] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of each of the above-described method embodiments.

[0164] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0165] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, and the like.

[0166] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0167] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A photovoltaic module defect detection method based on multi-physical quantity fusion, characterized in that: The method comprises: Obtain optical data of dust layer and dust weighing data of photovoltaic modules; Calculating a cleanliness compensation coefficient of the photovoltaic module according to the ash layer optical data and the dust weighing data; performing data correction on the original electroluminescent image of the photovoltaic module according to the cleanliness compensation coefficient to obtain a corrected electroluminescent image; inputting the corrected electroluminescent image into a defect cascade neural model of the photovoltaic module to obtain image defect analysis data; Defects of the photovoltaic module are delineated according to the image defect analysis data and physical constraints to obtain defect detection results of the photovoltaic module.

2. The method according to claim 1, characterized in that The calculating the cleanliness compensation coefficient of the photovoltaic module according to the ash layer optical data and the dust weighing data includes: Constructing a three-dimensional dust density distribution model based on the dust layer optical data and the dust weighing data; According to the light scattering physical model of the photovoltaic module, the three-dimensional distribution model of dust density is solved to obtain the cleanliness compensation coefficient of the photovoltaic module.

3. The method according to claim 2, characterized in that The step of constructing a three-dimensional dust density distribution model based on the dust layer optical data and the dust weighing data includes: constructing transmittance distribution data corresponding to the surface of the photovoltaic module according to the gray layer optical data; Calculating mass density information of the dust on the surface of the photovoltaic assembly based on the dust weighing data and the sensor sensing area of ​​the photovoltaic assembly; The transmittance distribution data and the mass density information are integrated to obtain the three-dimensional dust density distribution model.

4. The method according to claim 3, characterized in that The step of constructing transmittance distribution data corresponding to the surface of the photovoltaic module according to the gray layer optical data includes: Controlling the laser array corresponding to the photovoltaic assembly to scan the surface of the photovoltaic assembly to obtain transmitted light intensity distribution data of the photovoltaic assembly; Performing a pixel-level ratio operation on the transmitted light intensity distribution data and a dust-free reference image of the photovoltaic module to obtain a transmittance matrix; The transmittance matrix is ​​subjected to filtering and interpolation processing to generate the transmittance distribution data.

5. The method according to claim 3, characterized in that The calculating the mass density information of the dust on the surface of the photovoltaic assembly according to the dust weighing data and the sensor sensing area of ​​the photovoltaic assembly includes: Obtaining a total dust mass value of the photovoltaic module according to the dust mass perception of the photovoltaic module by the sensor sensing area; Dividing data sampling areas according to a spatial mapping relationship of the sensor sensing area on the surface of the photovoltaic module; According to the change rate of the transmittance distribution data in each of the data sampling areas, the total mass value of the dust is distributed to the surface of the photovoltaic component to obtain the mass density information.

6. The method according to claim 1, characterized in that The step of inputting the corrected electroluminescent image into a defect cascade neural model of the photovoltaic module to obtain image defect analysis data comprises: Performing component hidden crack defect detection on the corrected electroluminescent image to obtain component hidden crack detection data; Performing component broken gate defect detection on the corrected electroluminescent image to obtain component broken gate detection data; The component hidden crack detection data and the component broken grid detection data are collaboratively processed to obtain the image defect analysis data.

7. The method according to claim 6, characterized in that The performing component broken gate defect detection on the corrected electroluminescent image to obtain component broken gate detection data includes: Performing Kenny edge detection on the corrected electroluminescent image to obtain element gate line boundary information; Determine the position and direction of the element grid line boundary information according to the Hough transform algorithm to obtain candidate windows of each grid line area; Recognize the breakpoint position of each candidate window of the gate line area, and extract the length of each break gate and the corresponding break gate position; Each of the broken gate lengths and the corresponding broken gate positions are matched with the current density distribution model of the component to determine the broken gate detection data of the component.

8. The method according to claim 7, characterized in that The step of matching each of the broken gate lengths and the corresponding broken gate positions with a current density distribution model of a component to determine the broken gate detection data of the component includes: Calculating a two-dimensional current density distribution diagram of the photovoltaic module according to the design operating parameters and actual operating status information of the photovoltaic module; For any of the gate-breaking positions, calculating the angle between the gate line direction of the gate-breaking length and the mainstream current direction to obtain the gate-breaking direction angle; The component gate-break detection data is calculated according to the gate-break direction angle, the gate-break length, and the local current density.

9. The method according to claim 1, characterized in that Defining defects of the photovoltaic module based on the image defect analysis data and the physical constraint conditions to obtain defect detection results of the photovoltaic module includes: Verifying conflict information between the image defect analysis data and each sensor data according to the physical constraint condition; If the conflict information is not a null value, returning to the step of obtaining the optical data of the dust layer of the photovoltaic module and the dust weighing data until the conflict information is a null value; Based on the image defect analysis data, a confidence vote is performed on each preset defect of the photovoltaic module to obtain the defect detection result.

10. A photovoltaic module defect detection system based on multi-physical quantity fusion, characterized in that: The system includes a computer device and a terminal. The terminal is used to obtain optical data of the dust layer and dust weighing data of the photovoltaic module and transmit them to the computer device; The computer device is used to calculate the cleanliness compensation coefficient of the photovoltaic module based on the dust layer optical data and the dust weighing data; The computer device is used to perform data correction on the original electroluminescent image of the photovoltaic module according to the cleanliness compensation coefficient to obtain a corrected electroluminescent image; The computer device is configured to input the corrected electroluminescent image into a defect cascade neural model of the photovoltaic module to obtain image defect analysis data; The computer device is used to perform defect delineation on the photovoltaic module based on the image defect analysis data and physical constraints to obtain a defect detection result of the photovoltaic module.

Images