Photovoltaic panel detection method and system based on machine vision

Abstract

The invention relates to the technical field of target detection, in particular to a photovoltaic panel detection method and system based on machine vision, and the method comprises the following steps: obtaining the displacement data of a photovoltaic panel cell, calculating the displacement gradient of a connection point, dividing a rigid region and a non-rigid region, generating a deformation classification result, and constructing a global deformation field. Generating a time sequence deformation trend matrix; calculating a displacement mean value of a bracket fixing point; adjusting a deformation compensation parameter; generating an adaptive compensation parameter, adjusting a welding point position, calculating an abnormal region Euclidean distance, and generating a defect matching identification result. According to the method, rigid and non-rigid areas are accurately divided by calculating the displacement gradient of a connection point, a global deformation field is constructed in combination with the displacement change of a feature point, trend continuity is ensured, a compensation parameter is adjusted by utilizing a fixed point displacement mean value, a deformation increment factor is calculated, the weight and stability are corrected, the position of a welding point is adjusted, and inverse deformation correction is optimized. The Euclidean distance is combined to match defects, so that the detection precision and stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of target detection, and in particular to a photovoltaic panel detection method and system based on machine vision. Background Art

[0002] The field of target detection technology includes object recognition, positioning and classification based on computer vision and image processing technology. The core content of this field is to identify specific objects and determine their positions in the image by analyzing image or video data. Target detection technology usually consists of steps such as feature extraction, candidate region generation and classification. Feature extraction can be based on information such as edges, textures, colors, etc. Candidate region generation can be achieved through sliding windows, region generation networks, etc. Classification can be completed using traditional machine learning methods or deep learning models. This technology is widely used in many fields such as autonomous driving, security monitoring, medical image analysis and industrial inspection, and relies on the continuous development of methods such as deep neural networks, pattern recognition, computer vision and sensor fusion to improve detection accuracy and computational efficiency.

[0003] Among them, the photovoltaic panel detection method based on machine vision refers to the use of image analysis and target detection technology to identify and monitor the key components and operating status of electric vehicles. This method mainly monitors multiple technical matters such as battery status, body integrity, tire wear, light signals and dashboard display. Specific methods include abnormal identification of battery appearance based on image processing technology, damage detection of body structure using edge detection and deep learning models, assessment of tire wear using contour analysis and morphological processing methods, detection of light signal status combined with optical flow analysis and color recognition, and parsing of dashboard numerical information through character recognition and template matching technology.

[0004] Existing technologies are based on static image target detection and do not consider the deformation effects of photovoltaic panels caused by factors such as temperature and mechanical stress. They cannot effectively model the dynamic deformation process, which reduces the detection accuracy. Region division relies on edge and texture features and is easily affected by noise. The distinction between rigid and non-rigid regions is not accurate enough, which affects the accuracy of deformation compensation. Defect recognition is based on traditional image processing and does not combine deformation information. It is easy to misjudge defects due to slight deformations in the glass area, affecting recognition reliability. The adjustment of welding point position ignores the global deformation trend, resulting in the accumulation of compensation errors and affecting the overall correction effect. Deformation correction does not fully combine time series information, resulting in a lack of consistency in the correction results. Under the influence of long-term operation and complex environments, the detection stability and accuracy are difficult to guarantee. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a photovoltaic panel detection method and system based on machine vision.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a photovoltaic panel detection method based on machine vision, comprising the following steps: S1: Obtain the displacement data of multiple cells on the surface of the photovoltaic panel in the video, calculate the displacement vector gradient change rate of the photovoltaic cell connection point, divide the rigid area and the non-rigid area, and generate the photovoltaic panel area deformation classification result; S2: Based on the regional deformation classification result of the photovoltaic panel, the displacement change of the feature points between adjacent frames of the photovoltaic module encapsulation layer is calculated, the global deformation field of the photovoltaic panel is established, and the time series deformation trend matrix of the photovoltaic panel is generated; S3: Based on the photovoltaic panel time series deformation trend matrix, the overall displacement mean of the photovoltaic bracket fixing point is calculated, the deformation compensation parameters of the photovoltaic module encapsulation layer are adjusted, the deformation increment factor of the photovoltaic cell series connection area, the correction weight of the photovoltaic module non-rigid area and the deformation stability of the photovoltaic glass surface area are calculated, and the photovoltaic panel adaptive deformation compensation parameters are generated; S4: Based on the photovoltaic panel adaptive deformation compensation parameters, adjusting the positions of the characteristic points of the photovoltaic module welding points, calculating the mapping relationship, and generating a photovoltaic panel inverse deformation correction image; S5: Based on the photovoltaic panel inverse deformation correction image, the Euclidean distance of the morphological features between the abnormal area on the photovoltaic cell surface and the normal area on the photovoltaic cell surface is calculated to generate a photovoltaic panel defect matching recognition result.

[0007] The photovoltaic panel regional deformation classification results include the displacement vector gradient change rate of the rigid area, non-rigid area, and photovoltaic cell connection point; the photovoltaic panel time series deformation trend matrix includes the displacement change of the feature points between adjacent frames of the photovoltaic module encapsulation layer and the global deformation field of the photovoltaic panel; the photovoltaic panel adaptive deformation compensation parameters include the overall displacement mean of the photovoltaic bracket fixing point, the deformation compensation parameters of the photovoltaic module encapsulation layer, the deformation increment factor of the photovoltaic cell series connection area, the correction weight of the photovoltaic module non-rigid area, and the deformation stability of the photovoltaic glass surface area; the photovoltaic panel inverse deformation correction image includes the position adjustment and mapping relationship of the feature points of the photovoltaic module welding point; the photovoltaic panel defect matching and identification results include the Euclidean distance of the morphological features of the abnormal area on the surface of the photovoltaic cell and the Euclidean distance of the morphological features of the normal area of ​​the photovoltaic cell.

[0008] As a further solution of the present invention, the steps for obtaining the photovoltaic panel regional deformation classification result are specifically as follows: S101: Obtain displacement data of the solar cells on the surface of the photovoltaic panel, calculate the local displacement vector, obtain the gradient of the connection point displacement vector, and obtain the gradient distribution of the displacement vector; S102: calling the displacement vector gradient distribution, calculating the gradient value, screening high and low gradient areas, determining the distribution density, dividing the rigid and non-rigid areas, and obtaining the regional rigidity classification matrix; S103: calling the regional rigid classification matrix, analyzing the non-rigid regional connection points, calculating the deformation center, and calculating the deformation classification index in combination with the rigid regional distribution, using the formula: ; Calculate the regional deformation classification value of the photovoltaic panel, filter the deformation category, and obtain the regional deformation classification result of the photovoltaic panel; in, Represents the regional deformation classification value of the photovoltaic panel, Representative The displacement vector gradient of the non-rigid region, represents the distance from the region to the adjacent rigid region, Adjust the parameters for the deformation. Represents the first The deformation strength of a point, represents the average deformation strength of the rigid region adjacent to the point, represents the total number of non-rigid regions, Represents the total number of adjacent rigid regions.

[0009] As a further solution of the present invention, the steps for obtaining the photovoltaic panel time series deformation trend matrix are specifically as follows: S201: based on the photovoltaic panel regional deformation classification result, obtaining feature points between adjacent frames of the encapsulation layer, calculating the displacement change of the feature points, and obtaining a feature point displacement change matrix; S202: calling the feature point displacement change matrix, calculating the deformation vectors of multiple feature points in the global coordinate system, integrating local deformation information, and establishing a global deformation field of the photovoltaic panel; S203: calling the photovoltaic panel global deformation field, calculating the deformation trend values ​​of multiple time steps in the time series, using the formula: ; Calculate the time series deformation characteristics and establish the photovoltaic panel time series deformation trend matrix; in, Represents the total deformation index under the time series, Representative The influence coefficient of the next time step, Representative The displacement change in the time step is, Indicates The displacement deviation correction amount of the time step, Representative The sensitivity of the change in the time step, Smoothing parameters that represent deformation trends, Indicates The displacement normalization parameter, Represents the total number of time steps for the reference.

[0010] As a further solution of the present invention, the step of obtaining the photovoltaic panel adaptive deformation compensation parameters is specifically as follows: S301: Based on the photovoltaic panel time series deformation trend matrix, extract the displacement data of the photovoltaic bracket fixing point in the entire time series, and calculate the overall displacement mean of the photovoltaic bracket; S302: calling the overall displacement mean of the photovoltaic support, calculating the parameters of the photovoltaic module encapsulation layer, adjusting the parameters according to the compensation requirements of multiple module areas, and obtaining the photovoltaic module deformation compensation parameters; S303: Calling the photovoltaic module deformation compensation parameter, using the formula: ; Calculate the deformation increment factor of the photovoltaic cell series area; in, Represents the deformation increment factor of the photovoltaic cell series connection area, Representative The deformation change value of the series-connected area, represents the deformation weight of the region, Adjust the parameters for the deformation. Representing the region The deformation strength of the connection point, represents the mean deformation of the connection point, represents the total number of serially connected regions, The number of connection points representing the region; S304: calling the deformation increment factor of the photovoltaic cell series area, calculating the correction weight of the non-rigid area of ​​the photovoltaic module and the deformation stability of the photovoltaic glass surface, and establishing the photovoltaic panel adaptive deformation compensation parameters.

[0011] As a further solution of the present invention, the steps of acquiring the photovoltaic panel inverse deformation correction image are specifically as follows: S401: adjusting the position of the characteristic point of the photovoltaic module welding point based on the photovoltaic panel adaptive deformation compensation parameter, performing position correction according to the deformation compensation parameter, and obtaining the photovoltaic module welding point adjustment matrix; S402: calling the photovoltaic module welding point adjustment matrix, calculating the mapping relationship of the feature points before and after the adjustment, and establishing a coordinate transformation relationship for all adjustment points, using the formula: ; Calculate the mapping relationship between multiple feature points, obtain the spatial mapping matrix after deformation compensation, and obtain the photovoltaic module mapping relationship matrix; in, Representative Adjust the feature points to The mapping relationship value of the target feature points, Representative The feature point is The position offset in this adjustment, Representative The target feature point is The amount of correction in this adjustment, To adjust the offset correction factor, represents the normalization parameter, Representative The first adjustment The position of feature points in the original coordinates, Represents the new position of the same feature point after adjustment, represents the mapping smoothing parameter, Representative The adjusted feature points are The coordinate values ​​in the round transformation, Representative The coordinate values ​​of the target feature points in the same round of transformation, Represents the number of adjustments, Represents the total number of feature points, Represents the total number of transformation operations; S403: calling the photovoltaic assembly mapping relationship matrix, applying transformation parameters to adjust the image pixel coordinates, reconstructing the corrected pixel distribution, performing image interpolation to fill the pixel gaps, and generating an inverse deformation correction image of the photovoltaic panel.

[0012] As a further solution of the present invention, the steps of obtaining the photovoltaic panel defect matching and identification results are specifically as follows: S501: based on the photovoltaic panel inverse deformation correction image, obtaining morphological features of abnormal areas and normal areas on the surface of the photovoltaic cell, extracting key morphological parameters, and obtaining a photovoltaic cell area morphological feature set; S502: calling the photovoltaic cell area morphological feature set, calculating the Euclidean distance of the morphological features between the abnormal area and the normal area, and performing distance weighted operations on multiple morphological feature dimensions, using the formula: ; Calculate the Euclidean distance of morphological features, screen the feature matching degree of abnormal areas, and obtain the Euclidean distance matrix of photovoltaic panel morphological features; in, Represents abnormal area With normal area The Euclidean distance of the morphological features between Representative The weighting factor of the morphological features, Represents abnormal area In the The value of the morphological feature dimension, Represents the normal area The values ​​on the same feature dimension, represents the norm parameter of the morphological feature distance calculation, represents the feature matching adjustment parameter, Represents abnormal area In the The values ​​on the auxiliary feature dimensions, Represents the normal area The values ​​on the same auxiliary feature dimension, Represents the total number of key features, Represents the total number of auxiliary features; S503: calling the Euclidean distance matrix of the photovoltaic panel morphological features, matching the morphological feature relationship between the abnormal area and the normal area, and establishing a photovoltaic panel defect matching and recognition result.

[0013] A photovoltaic panel detection system based on machine vision, the photovoltaic panel detection system based on machine vision is used to perform the photovoltaic panel detection method based on machine vision, the system comprises: The displacement data calculation module obtains the displacement data of the solar cells on the surface of the photovoltaic panel, calls the connection point coordinate information, calculates the displacement vector gradient change rate, selects points below the threshold to divide the rigid area, selects points above the threshold to divide the non-rigid area, integrates the rigid and non-rigid area distribution, and obtains the photovoltaic panel area deformation classification result; The deformation trend analysis module calls the coordinates of the feature points of the adjacent frames of the encapsulation layer based on the deformation classification results of the photovoltaic panel area, calculates the displacement change, integrates the global deformation field, and obtains the photovoltaic panel time series deformation trend matrix; The deformation compensation calculation module calls the displacement of the bracket fixed point based on the photovoltaic panel time series deformation trend matrix, calculates the overall mean, adjusts the compensation parameters, calculates the deformation increment factor, and obtains the photovoltaic panel adaptive deformation compensation parameters; The inverse deformation correction module adjusts the position of the characteristic points of the welding points based on the adaptive deformation compensation parameters of the photovoltaic panel, calculates the mapping relationship, and obtains the inverse deformation correction image of the photovoltaic panel; The defect recognition module calculates the Euclidean distance between the abnormal area and the normal area based on the inverse deformation correction image of the photovoltaic panel, matches the defect feature library, and obtains the photovoltaic panel defect matching recognition result.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by calculating the displacement vector gradient change rate of multiple cell connection points of the photovoltaic panel, the rigid and non-rigid areas are finely divided to make the deformation classification more accurate. Combined with the displacement changes of the feature points between adjacent frames of the encapsulation layer, a global deformation field is established to make the deformation trend have time continuity. The deformation compensation parameters are adjusted using the overall displacement mean of the fixed point to adapt the compensation to the characteristics of different regions. The deformation increment factor of the series connection area, the correction weight of the non-rigid area and the deformation stability of the glass surface area are calculated to make the compensation more targeted. The position of the characteristic points of the welding point is adjusted through the mapping relationship to achieve a more stable inverse deformation correction. When matching defects, the Euclidean distance of the morphological features of the abnormal and normal areas is calculated so that the recognition process is combined with deformation information to improve accuracy. The overall solution combines dynamic deformation analysis, global deformation trend modeling and precise deformation compensation to ensure that photovoltaic panel detection maintains high accuracy and stability in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the workflow of the present invention; Figure 2 A flowchart of the steps for obtaining the regional deformation classification results of the photovoltaic panel of the present invention; Figure 3 A flow chart of the steps for obtaining the time series deformation trend matrix of the photovoltaic panel of the present invention; Figure 4 A flow chart of the steps for obtaining the adaptive deformation compensation parameters of the photovoltaic panel of the present invention; Figure 5 A flow chart of the steps for obtaining the reverse deformation correction image of the photovoltaic panel of the present invention; Figure 6 This is a flow chart of the steps for obtaining the photovoltaic panel defect matching and identification results of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0018] Embodiment 1 See also Figure 1 The present invention provides a technical solution: a photovoltaic panel detection method based on machine vision, comprising the following steps: S1: Obtain the displacement data of multiple cells on the surface of the photovoltaic panel in the video, calculate the displacement vector gradient change rate of the photovoltaic cell connection point, divide the rigid area and the non-rigid area, and generate the photovoltaic panel area deformation classification result; S2: Based on the regional deformation classification results of the photovoltaic panel, the displacement change of the feature points between adjacent frames of the photovoltaic module encapsulation layer is calculated, the global deformation field of the photovoltaic panel is established, and the deformation trend matrix of the photovoltaic panel time series is generated; S3: Based on the deformation trend matrix of the photovoltaic panel time series, the overall displacement mean of the photovoltaic bracket fixing point is calculated, the deformation compensation parameters of the photovoltaic module encapsulation layer are adjusted, the deformation increment factor of the photovoltaic cell series connection area, the correction weight of the non-rigid area of ​​the photovoltaic module and the deformation stability of the photovoltaic glass surface area are calculated, and the photovoltaic panel adaptive deformation compensation parameters are generated; S4: Based on the photovoltaic panel adaptive deformation compensation parameters, adjust the positions of the characteristic points of the photovoltaic module welding points, calculate the mapping relationship, and generate the photovoltaic panel inverse deformation correction image; S5: Based on the inverse deformation correction image of the photovoltaic panel, the Euclidean distance of the morphological features between the abnormal area on the surface of the photovoltaic cell and the normal area on the photovoltaic cell is calculated to generate a photovoltaic panel defect matching recognition result.

[0019] The regional deformation classification results of photovoltaic panels include the displacement vector gradient change rate of rigid areas, non-rigid areas, and photovoltaic cell connection points. The photovoltaic panel time series deformation trend matrix includes the displacement change of feature points between adjacent frames of the photovoltaic module encapsulation layer and the global deformation field of the photovoltaic panel. The adaptive deformation compensation parameters of the photovoltaic panel include the overall displacement mean of the photovoltaic bracket fixing point, the deformation compensation parameters of the photovoltaic module encapsulation layer, the deformation increment factor of the photovoltaic cell series connection area, the correction weight of the non-rigid area of ​​the photovoltaic module, and the deformation stability of the photovoltaic glass surface area. The photovoltaic panel inverse deformation correction image includes the position adjustment and mapping relationship of the feature points of the photovoltaic module welding points. The photovoltaic panel defect matching and recognition results include the Euclidean distance of the morphological features of the abnormal area on the surface of the photovoltaic cell and the Euclidean distance of the morphological features of the normal area of ​​the photovoltaic cell.

[0020] See also Figure 2 , the specific steps for obtaining the regional deformation classification results of photovoltaic panels are: S101: Obtain displacement data of the solar cells on the surface of the photovoltaic panel, calculate the local displacement vector, obtain the gradient of the connection point displacement vector, and obtain the gradient distribution of the displacement vector; It is necessary to arrange high-precision displacement sensors on different photovoltaic modules. Each sensor is installed on the surface of the cell in a uniform distribution manner and records the displacement of the cell in real time under changes in the external environment (temperature, wind speed, light, etc.). For example, 4 sensors are arranged on the surface of each cell to record the displacement information of its four corner points respectively, and collect data at a time interval of 1s; these data are transmitted to the calculation unit through the data acquisition module, and the calculation unit calculates the data of each monitoring point of each cell to obtain the local displacement vector. For example, for a certain cell, the displacements recorded at its four corner points are mm, then the local displacement vector of the cell can be calculated by the average displacement of the four points, and its value is mm; based on the local displacement vector of each cell, the displacement vectors of all adjacent cells are calculated, and the central difference method is used to obtain the displacement vector gradient of each connection point. Specifically, if the displacement vectors of two adjacent cells are 0.13 mm and 0.09 mm respectively, the gradient calculation method is: ,in is the cell spacing, assuming mm, then the calculated gradient value is mm / mm; the gradients of all connection points on the photovoltaic panel are calculated in the same way to obtain the displacement vector gradient distribution.

[0021] S102: calling the displacement vector gradient distribution, calculating the gradient value, screening the high and low gradient areas, determining the distribution density, dividing the rigid and non-rigid areas, and obtaining the regional rigidity classification matrix; First, calculate the mean gradient of the entire panel area , assuming that the gradient values ​​of all connection points are: ; Then calculate the mean: mm / mm; Then compare the gradient value of each connection point with the mean value and set the judgment threshold , the threshold is generally 1.2 times the mean, that is, mm / mm. If the gradient value of a connection point is higher than the threshold, it is classified as a high gradient area, otherwise it is classified as a low gradient area. For example, if the gradient values ​​of the connection points are 0.02, 0.03, 0.05, 0.01, and 0.04, 0.05 and 0.04 are classified as high gradient areas, and the rest are classified as low gradient areas. Then, the distribution density of high gradient areas and low gradient areas is calculated. The density calculation method is the number of high gradient points per unit area. Assuming that the area of ​​a certain area is 10 , which contains 3 high gradient points, then the high gradient distribution density of this area is / cm2; all areas are divided into The regions with density higher than the threshold are classified as non-rigid regions, and those with density lower than the threshold are classified as rigid regions, and the regional rigidity classification matrix is ​​obtained.

[0022] S103: calling the regional rigid classification matrix, analyzing the non-rigid regional connection points, calculating the deformation center, and calculating the deformation classification index in combination with the rigid regional distribution, using the formula: ; Calculate the regional deformation classification value of the photovoltaic panel, filter the deformation category, and obtain the regional deformation classification result of the photovoltaic panel; in, Represents the regional deformation classification value of the photovoltaic panel, Representative The displacement vector gradient of the non-rigid region, represents the distance from the region to the adjacent rigid region, Adjust the parameters for the deformation. Represents the first The deformation strength of a point, represents the average deformation strength of the rigid region adjacent to the point, represents the total number of non-rigid regions, Represents the total number of adjacent rigid regions.

[0023] First, calculate the deformation center of the non-rigid area. It is calculated by the centroid of all connection points in the non-rigid region. Assuming that a non-rigid region contains 4 connection points with coordinates (1,2), (3,4), (5,6), and (7,8), the deformation center is calculated as , ; That is, the coordinates of the deformation center are (4,5); then, the minimum distance from the deformation center to the adjacent rigid area is calculated. Assuming that the coordinates of the nearest connection point of the adjacent rigid area are (8,10), the distance between the two points is calculated as the Euclidean distance: mm; Next, calculate the deformation classification index using the formula: ; Among them, , the non-rigid region contains 3 connection points, and their gradients are mm / mm, the distance values ​​are mm, and the deformation strengths of adjacent rigid regions are mm / mm, and the deformation strengths of the non-rigid regions are mm / mm, the deformation classification index is calculated as follows: 1. 2. 3. 4. Calculate the deformation classification index value ,If the deformation classification index is greater than the threshold value of 0.75, it is determined that the area has undergone obvious deformation, so the area is classified as a deformation area, and finally the deformation classification result of the photovoltaic panel area is obtained.

[0024] See also Figure 3 , the specific steps for obtaining the time series deformation trend matrix of photovoltaic panels are: S201: based on the regional deformation classification result of the photovoltaic panel, obtaining feature points between adjacent frames of the encapsulation layer, calculating the displacement change of the feature points, and obtaining a feature point displacement change matrix; When photovoltaic panels are operated in outdoor environments for a long time, their packaging layer will be affected by changes in the external environment (temperature, wind, light intensity, etc.), causing deformation, resulting in displacement differences between the characteristic points of the photovoltaic components in adjacent time frames. First, the positions of several characteristic points on the surface of the photovoltaic components can be determined. High-precision camera equipment can be used to collect images, and computer vision processing technology can be used to extract the coordinate information of the characteristic points. For example, the optical flow method is used to extract the coordinates of the characteristic points between time frames. The time interval is set to 0.1s. If a certain characteristic point is in the first The coordinates of the frame are ( ), in The coordinates of the frame are ( ), then the displacement change of the feature point is calculated as follows: ; For example, for a feature point, its The frame coordinates are (102.3, 205.7). The coordinates of the frame are (104.8, 207.2), then calculate its displacement change: ; This calculation requires batch calculation of all feature points on the surface of the PV module encapsulation layer to form an overall displacement change matrix, as shown in Table 1.

[0025] Table 1. Matrix of displacement changes of characteristic points of photovoltaic panels As shown in Table 1, the displacement change of each feature point is calculated, and the displacement change matrix of all feature points is combined to obtain the feature point displacement change matrix.

[0026] S202: calling the feature point displacement change matrix, calculating the deformation vectors of multiple feature points in the global coordinate system, integrating the local deformation information, and establishing the global deformation field of the photovoltaic panel; Call the feature point displacement change matrix, calculate the deformation vector of each feature point in the global coordinate system, integrate the local deformation information, and use the two-dimensional coordinate transformation method to project the local displacement of the feature point into the global deformation field. First, set the origin position of the global coordinate system, and take the center point of the photovoltaic module as the global coordinate origin ( ), then the local displacement vector of the feature point is expressed as: ; in, , , for example, the coordinate change of a feature point is (104.8, 207.2) → (107.3, 209.6), then: ; Calculate the deformation vector of the feature point: ; By accumulating and normalizing the deformation vectors of all feature points, a global deformation field is established, as shown in Table 2.

[0027] Table 2 Global deformation field matrix As shown in Table 2, the local deformation vector of each feature point has been mapped to the global coordinate system, and finally the global deformation field of the photovoltaic panel is established.

[0028] S203: calling the photovoltaic panel global deformation field, calculating the deformation trend values ​​of multiple time steps in the time series, using the formula: ; Calculate the time series deformation characteristics and establish the photovoltaic panel time series deformation trend matrix; in, Represents the total deformation index under the time series, Representative The influence coefficient of the next time step, Representative The displacement change in the time step is, Indicates The displacement deviation correction amount of the time step, Representative The sensitivity of the change in the time step, Smoothing parameters that represent deformation trends, Indicates The displacement normalization parameter, Represents the total number of time steps for the reference.

[0029] formula: ; Calculate the deformation trend index of each time step, set the time interval to 0.1s, and assume that the deformation monitoring time window of a photovoltaic panel is 10s, that is, , given data as follows: Influence coefficient ; Displacement change ; Deviation correction amount ; Change sensitivity ; Adjusting parameters ; Standardization parameters ; Bring it into calculation: ; Calculate the numerator part: ; Calculate the denominator: ; Final calculation: ; The results show that in the current time series window, the deformation trend index of the photovoltaic panel is 4.12. This value can be used to monitor the overall deformation degree of the photovoltaic panel during operation. Combined with historical data analysis, it can further determine whether the deformation trend is within the normal range, and finally establish the photovoltaic panel time series deformation trend matrix.

[0030] See also Figure 4 , the specific steps for obtaining the adaptive deformation compensation parameters of photovoltaic panels are as follows: S301: Based on the photovoltaic panel time series deformation trend matrix, the displacement data of the photovoltaic bracket fixing point in the entire time series is extracted, and the overall displacement mean of the photovoltaic bracket is calculated; First, the displacement data of the fixed points of the photovoltaic bracket in the entire time series are extracted. During the extraction process, multiple fixed monitoring points are set, and each monitoring point needs to record the displacement changes at a series of time points. The time interval is set to 1 hour, 24 records are recorded every day, and 30 days of data are accumulated to form a long-term trend matrix. For the displacement data of each monitoring point, it is necessary to calculate its displacement increment, that is, the difference between the displacement value at the current time point and the displacement value at the previous time point, to form a time series displacement change curve. When calculating the overall displacement mean of the photovoltaic bracket, the displacement data of all fixed points are first summed and then divided by the total number of fixed points to obtain the single-day average, and further calculate the 30-day average trend. The average trend is calculated using the sliding average method, and the window length is set to 7 days to reduce short-term data fluctuations. The selection of the window length is based on the fluctuation range of historical monitoring data, that is, if the standard deviation of the 7-day sliding mean is less than 0.02 mm, the 7-day window is maintained, otherwise it is expanded to 10 days. If the displacement increment of a fixed point exceeds 1.5 times the overall mean, it is marked as an abnormal point, and the validity of its data source is re-evaluated. The setting of the 1.5-fold threshold is based on the deformation limit of the photovoltaic bracket material. For example, the allowable elastic deformation of the aluminum alloy bracket generally does not exceed 0.3 mm. Therefore, the maximum fluctuation range of the bracket without external force is calculated through historical data, and the average fluctuation upper limit is about 0.2 mm. After taking 1.5 times, it is set to 0.3 mm to ensure that the threshold can identify abnormal data without accidentally damaging normal fluctuations. The overall displacement mean of the photovoltaic bracket is calculated through the screened data, as shown in Table 3.

[0031] Table 3 Monitoring data of fixed point displacement of photovoltaic bracket As shown in Table 3, through monitoring of multi-day data, it can be found that the average overall displacement of the photovoltaic bracket is about 0.13mm. This data will be used for subsequent calculation of the adjustment of the photovoltaic module encapsulation layer parameters.

[0032] S302: calling the overall displacement mean of the photovoltaic bracket, calculating the parameters of the photovoltaic module encapsulation layer, adjusting the parameters according to the compensation requirements of multiple module areas, and obtaining the deformation compensation parameters of the photovoltaic module; The overall displacement mean of the photovoltaic bracket is called, and the calculation is performed based on the compensation demand for component area deformation. First, for the encapsulation layer of the photovoltaic component, the deformation of the component area needs to be obtained. The measurement interval is set to 1 hour, and the surface displacement of each component area is recorded. The mean of the component area deformation is obtained through matrix calculation, and the difference is calculated with the overall displacement mean of the photovoltaic bracket to obtain the deformation adjustment value of the component encapsulation layer. If the deformation of a component area exceeds the overall mean by more than 20%, the compensation weight needs to be increased. The 20% threshold is set based on the allowable stress distribution range of the photovoltaic encapsulation layer. According to material mechanics calculations, the thermal expansion rate of the photovoltaic encapsulation layer under the influence of ambient temperature difference is about 0.04%. Considering the daily temperature difference of 30°C, the maximum The large deformation change rate is about 1.2%, so 20% is the optimized value after empirical adjustment. The weight is calculated using the proportional correction method. The weight setting range is 1.0-2.5, and the benchmark value is set to 1.5. The benchmark value is set based on the influence of the component regional deformation on the power generation efficiency without compensation. If the deformation variable is greater than 30% of the overall mean, the component power attenuation is about 2.3%. Therefore, the benchmark value of 1.5 can moderately reduce the deformation impact. If the deformation variable is less than 5% of the mean, the weight is set to 1.0. If the deformation variable exceeds the mean by 30%, it is set to 2.5. When calculating the deformation compensation parameters of the photovoltaic module encapsulation layer, the final compensation parameters are calculated based on the deformation adjustment values ​​and weights of each region, as shown in Table 4.

[0033] Table 4 Calculation of deformation compensation parameters of photovoltaic module encapsulation layer S303: Calling the photovoltaic module deformation compensation parameter, using the formula: ; Calculate the deformation increment factor of the photovoltaic cell series area; in, Represents the deformation increment factor of the photovoltaic cell series connection area, Representative The deformation change value of the series-connected area, represents the deformation weight of the region, Adjust the parameters for the deformation. Representing the region The deformation strength of the connection point, represents the mean deformation of the connection point, represents the total number of serially connected regions, The number of connection points representing the region; formula: ; In this calculation process, the deformation change value of the series connection area must first be determined , which is derived from the displacement change of the photovoltaic module area and is obtained by differential calculation, namely: ; in and are the deformation value and deformation weight of the current moment and the previous moment respectively. According to the regional influence, the value range is 0.8-2.0, adjust the parameters According to the regional connection strength setting, the value range is 0.5-1.5, the connection point deformation strength The deformation mean is obtained through monitoring data It is calculated by taking the average of the deformation values ​​of all connection points; In the process of calculating the deformation increment factor of the photovoltaic cell series area, the deformation change value is first determined , using the differential calculation method, determined by the difference in deformation between the current moment and the previous moment. In this example, , ,get . Deformation Weight Set according to the regional impact, ranging from 0.8-2.0, in the example . Adjust the parameters Reflects the regional connection strength, ranging from 0.5-1.5, taken in the example . Connection point deformation strength Obtained from monitoring data, deformation mean Calculated by the deformation values ​​of all connection points, in this example, , .

[0034] Bring in parameters: ; ; ; The deformation increment factor of the photovoltaic cell series area is calculated .

[0035] S304: calling the deformation increment factor of the photovoltaic cell series area, calculating the correction weight of the non-rigid area of ​​the photovoltaic module and the deformation stability of the photovoltaic glass surface, and establishing the photovoltaic panel adaptive deformation compensation parameters.

[0036] The correction weight of the non-rigid area is set according to the size of the deformation increment factor. The setting of the correction weight refers to the influence of component power attenuation under different deformation conditions. , the correction weight is set to 0.9, if , the correction weight is set to 1.0, if , the correction weight is set to 1.2, and the deformation stability of the photovoltaic glass surface is calculated at the same time. The stability calculation is based on the standard deviation analysis of the 30-day deformation mean. If the standard deviation of the daily average deformation is less than 0.02mm, the deformation stability is set to high. If the standard deviation is between 0.02-0.05mm, the deformation stability is set to medium. If the standard deviation is greater than 0.05mm, the deformation stability is set to low. In this calculation, the deformation standard deviation is 0.03mm, so the deformation stability is set to medium. Finally, based on the correction weights of each area, the adaptive deformation compensation parameters of the photovoltaic panel are established.

[0037] See also Figure 5 , the specific steps for obtaining the inverse deformation correction image of the photovoltaic panel are: S401: adjusting the position of the characteristic point of the photovoltaic module welding point based on the photovoltaic panel adaptive deformation compensation parameter, performing position correction according to the deformation compensation parameter, and obtaining the photovoltaic module welding point adjustment matrix; For each welding point feature point, image analysis technology is used to identify its pixel position and convert it to the actual physical coordinate system. Assuming that the initial coordinates of a welding point are (120.5mm, 80.3mm), combined with the photovoltaic panel adaptive deformation compensation parameters, the deformation compensation vector of the point can be expressed as (Δx=2.1mm, Δy=-1.5mm), and the corrected coordinates of the point can be calculated as (122.6mm, 78.8mm). The same calculation is performed on the feature points of all welding points, and a table is used to record the position information before and after compensation, as shown in Table 5.

[0038] Table 5 Welding point feature point position adjustment data table As shown in Table 5, the coordinates of each feature point were corrected after deformation compensation, and the corrected coordinate data of all welding points were called to construct the photovoltaic module welding point adjustment matrix.

[0039] S402: calling the photovoltaic module welding point adjustment matrix, calculating the mapping relationship of the feature points before and after the adjustment, and establishing a coordinate transformation relationship for all adjustment points, using the formula: ; Calculate the mapping relationship between multiple feature points, obtain the spatial mapping matrix after deformation compensation, and obtain the photovoltaic module mapping relationship matrix; in, Representative Adjust the feature points to The mapping relationship value of the target feature points, Representative The feature point is The position offset in this adjustment, Representative The target feature point is The amount of correction in this adjustment, To adjust the offset correction factor, represents the normalization parameter, Representative The first adjustment The position of feature points in the original coordinates, Represents the new position of the same feature point after adjustment, represents the mapping smoothing parameter, Representative The adjusted feature points are The coordinate value in the round transformation, Representative The coordinate values ​​of the target feature points in the same round of transformation, Represents the number of adjustments, Represents the total number of feature points, Represents the total number of transformation operations; Call the photovoltaic module welding point adjustment matrix, calculate the mapping relationship of the feature points before and after the adjustment, and select the welding point feature point set {P1, P2, …, Pn}, where the coordinates of each point before and after the adjustment are and , when calculating the mapping relationship, the deformation compensation vector and space mapping matrix of the welding point need to be considered, and the formula is: ; The mapping relationship between multiple feature points is calculated, the spatial mapping matrix after deformation compensation is obtained, and the mapping relationship matrix of the photovoltaic module is obtained.

[0040] Calculate the parameter assignment in the formula: set up mm, mm; Pick , , , ; set up , , ; calculate: Calculate an element of the photovoltaic module mapping relationship matrix ,The result shows that the mapping relationship of the welding points after ,adjustment is closer, and the spatial mapping matrix after ,deformation compensation can accurately describe the position relationship of the ,feature points after adjustment.

[0041] S403: calling the photovoltaic component mapping relationship matrix, applying the transformation parameters to adjust the image pixel coordinates, reconstructing the corrected pixel distribution, performing image interpolation to fill the pixel gaps, and generating the photovoltaic panel inverse deformation correction image.

[0042] First, it is necessary to obtain the projection area of ​​the photovoltaic component in the image. The projection area is determined by factors such as the geometric shape, installation angle, and lighting direction of the photovoltaic component. Specifically, it is necessary to extract the actual position information of the photovoltaic component in the three-dimensional coordinate system, and calculate its projection boundary in the image coordinate system based on the optical parameters of the sensor (including focal length, principal point coordinates, distortion coefficient, etc.). Subsequently, a one-to-one correspondence is established between the obtained projection area of ​​the photovoltaic component and the pixel coordinates in the image. This process involves coordinate transformation. The spatial coordinates of the photovoltaic component are represented by homogeneous coordinates, which are transformed into perspective projection through the external parameter matrix and the internal parameter matrix to calculate the projection point in the image coordinates. If the photovoltaic component has a tilt angle, the projection distortion needs to be further considered and corrected using affine transformation. The transformation matrix involved in this process can be corrected by measured data. For example, by selecting multiple feature points in the image and combining them with the actual installation data of the photovoltaic component, a one-to-one correspondence is established between the obtained projection area of ​​the photovoltaic component and the pixel coordinates in the image. The transformation matrix is ​​obtained by least squares optimization. The specific optimization process is as follows: 1) Select at least 4 known matching point pairs, 2) Construct an error function, which represents the square error between the transformed point and the actual point, 3) Calculate the partial derivatives of the error with respect to each parameter of the transformation matrix, 4) Use gradient descent to iteratively update the parameters until they converge to within the threshold ε=0.001, and obtain the final mapping relationship matrix. After completing the image coordinate adjustment, reconstruct the corrected pixel distribution. This step needs to ensure the integrity of the pixels, that is, all adjusted pixels must still be in the effective area of ​​the image, and at the same time, ensure that the pixel grayscale value is within a reasonable range (0-255). If pixels overlap, calculate the average value of the overlapping pixels for fusion. If some pixels are missing due to transformation, image interpolation is required to fill the pixel gaps. Bilinear interpolation is used here, that is, for missing pixels, their values ​​are calculated by weighted average of the grayscale values ​​of the four surrounding known pixels. The calculation formula is: ; in, is the gray value of the missing pixel, are the grayscale values ​​of the four surrounding known pixels, and are the normalized position parameters of the pixel in the horizontal and vertical directions. For example, if a missing pixel is at coordinates (12.3,45.7), and its four adjacent known pixels are (12,45), (13,45), (12,46), and (13,46), then , substitute into the formula to calculate the gray value of the missing point, and generate the photovoltaic panel inverse deformation correction image, which is used for subsequent analysis and processing.

[0043] See also Figure 6 , the specific steps for obtaining the photovoltaic panel defect matching and identification results are as follows: S501: based on the photovoltaic panel inverse deformation correction image, the morphological features of the abnormal area and the normal area on the surface of the photovoltaic cell are obtained, key morphological parameters are extracted, and a set of morphological features of the photovoltaic cell area is obtained; First, several key areas in the image are selected as reference points. The selection of reference points can be based on parameters such as the uniformity of the cell surface texture, the consistency of color distribution, and the characteristics of light reflection. For example, among 100 photovoltaic cell samples, there may be 10-15 abnormal areas, and the brightness value distribution of these areas is different. When extracting features, it is necessary to quantify the edge contour, grayscale gradient change, and area morphology parameters of each area, and set the edge sharpness threshold of the abnormal area to between 50 and 100. When the sharpness value is higher than this range, there may be cracks or foreign objects blocking the edge. The threshold is set based on the surface detection data of standard photovoltaic modules. Specifically, By measuring the sharpness distribution of 500 normal photovoltaic cells and 500 abnormal cells, the normal range is between 40-90, while the abnormal area is usually between 55-120, so 50-100 is selected as the detection range. By calculating the morphological parameters such as the major axis, minor axis ratio, eccentricity and boundary connectivity of each area, its classification can be preliminarily judged, and then statistical normalization processing is performed to facilitate further matching analysis in subsequent steps. The normalization range of the morphological parameters is set to [0,1]. The normalization calculation adopts the minimum-maximum standardization method, and the normalization reference value is set to the range of the minimum and maximum values ​​of the normal photovoltaic module characteristics. That is, the normalization formula is: ; For example, if the area parameter of an abnormal area is 1200px2, and the area distribution range of the normal area is 800-2000px2, the normalized value is: ; All calculated morphological characteristic parameters are summarized to obtain the photovoltaic cell regional morphological characteristic set.

[0044] S502: Calling the photovoltaic cell area morphological feature set, calculating the Euclidean distance of the morphological features between the abnormal area and the normal area, and performing distance weighted operations on multiple morphological feature dimensions, using the formula: ; Calculate the Euclidean distance of morphological features, screen the feature matching degree of abnormal areas, and obtain the Euclidean distance matrix of photovoltaic panel morphological features; in, Represents abnormal area With normal area The Euclidean distance of the morphological features between Representative The weighting factor of the morphological features, Represents abnormal area In the The value of the morphological feature dimension, Represents the normal area The values ​​on the same feature dimension, represents the norm parameter of the morphological feature distance calculation, represents the feature matching adjustment parameter, Represents abnormal area In the The values ​​on the auxiliary feature dimensions, Represents the normal area The values ​​on the same auxiliary feature dimension, Represents the total number of key features, Represents the total number of auxiliary features; In the process of calculating the morphological features of photovoltaic cells, it is necessary to calculate the Euclidean distance of the morphological features between the abnormal area and the normal area. To achieve this calculation, it is first necessary to obtain the morphological feature data sets of the abnormal area and the normal area. The morphological feature data of the abnormal area can be extracted by image processing methods. For example, after the edge detection algorithm is used to determine the area contour, the morphological method is used to calculate its basic morphological parameters such as area, perimeter, aspect ratio, etc. The morphological feature data of the normal area can be extracted from a large number of normal samples by statistical methods, and a reference database is established. After the data extraction is completed, the morphological feature data of the abnormal area and the normal area need to be matched in order to subsequently calculate their Euclidean distance. In the calculation process, the difference of each morphological feature dimension needs to be weighted and the weighting factor of each morphological feature needs to be set. The weight value can be set according to the importance of the feature to defect recognition. For example, area and aspect ratio may have higher distinguishing ability, so they are given a larger weight, while minor features such as edge complexity can be set with a lower weight. For the norm parameter of morphological feature distance calculation , whose value determines the calculation method, for example corresponds to the standard Euclidean distance, and When corresponds to the Manhattan distance, the parameter The setting of will affect the sensitivity of matching abnormal areas with normal areas. After calculating the morphological feature distance, auxiliary feature matching should also be considered. For example, color, texture, etc. can be used as auxiliary features to improve matching accuracy. These auxiliary features are calculated through Euclidean distance and the parameters are used. Control its influence weight on the overall matching, assuming that an abnormal area The morphological characteristics data are , normal area The morphological characteristics data are , the corresponding weight factor is set as , the auxiliary feature data is and , and set the auxiliary feature weight The calculation is based on the following formula: ; Assume the following specific values: Table 6 Morphological characteristics data table As shown in Table 6, the values ​​of different morphological features in abnormal areas and normal areas and their corresponding weights are listed. The weight factor is used to adjust the impact of each feature on the matching calculation.

[0045] Table 7 Auxiliary feature data table As shown in Table 7, the values ​​of auxiliary features in abnormal areas and normal areas are listed, and these features will be taken into consideration when calculating the Euclidean distance.

[0046] Calculate the morphological feature distance: ; ; ; Calculate auxiliary feature distance: ; ; Final Euclidean distance: ; The results show that the abnormal area and normal area The Euclidean distance of the morphological characteristics between them is 0.637, which can be used to further classify whether they belong to the same type of area.

[0047] S503: calling the Euclidean distance matrix of photovoltaic panel morphological features, matching the morphological feature relationship between the abnormal area and the normal area, and establishing photovoltaic panel defect matching and recognition results.

[0048] Call the Euclidean distance matrix of photovoltaic panel morphological features to match the morphological feature relationship between abnormal areas and normal areas. In order to match, it is necessary to sort the data in the Euclidean distance matrix and set the matching threshold. , the value is set to 0.08, which is based on the matching error analysis of 5000 sets of photovoltaic cell defect samples. Under different threshold settings, the error rate curve shows that when When the value is 0.08, the mean of the false positive rate and the false negative rate is the lowest. Assuming that there are 5 abnormal areas in a photovoltaic cell, they are matched with 10 normal areas respectively. The calculated Euclidean distance is shown in Table 8: Table 8 Euclidean distance matching table of morphological features As shown in Table 8, based on the calculation results of the Euclidean distance of morphological features, the matching between abnormal areas and normal areas is as follows: area A matches normal areas 1 and 7 (because its Euclidean distance is lower), area B matches normal areas 3 and 5, area C matches normal areas 4 and 5, area D matches normal areas 2, 3, and 7, and area E matches normal areas 4, 5, and 6.

[0049] Finally, the matching relationship between the defective area and the normal area of ​​the photovoltaic panel is established through the calculation of the Euclidean distance of the morphological features, providing a reference for subsequent defect identification.

[0050] A photovoltaic panel detection system based on machine vision, the photovoltaic panel detection system based on machine vision is used to execute the photovoltaic panel detection method based on machine vision, the system comprises: The displacement data calculation module obtains the displacement data of the solar cells on the surface of the photovoltaic panel, calls the connection point coordinate information, calculates the displacement vector gradient change rate, selects points below the threshold to divide the rigid area, selects points above the threshold to divide the non-rigid area, integrates the rigid and non-rigid area distribution, and obtains the photovoltaic panel area deformation classification result; The deformation trend analysis module is based on the regional deformation classification results of the photovoltaic panel, calls the coordinates of the feature points of the adjacent frames of the encapsulation layer, calculates the displacement change, integrates the global deformation field, and obtains the time series deformation trend matrix of the photovoltaic panel; The deformation compensation calculation module is based on the photovoltaic panel time series deformation trend matrix, calls the bracket fixed point displacement, calculates the overall mean, adjusts the compensation parameters, calculates the deformation increment factor, and obtains the photovoltaic panel adaptive deformation compensation parameters; The reverse deformation correction module adjusts the position of the characteristic points of the welding points based on the adaptive deformation compensation parameters of the photovoltaic panel, calculates the mapping relationship, and obtains the reverse deformation correction image of the photovoltaic panel; The defect recognition module calculates the Euclidean distance between the abnormal area and the normal area based on the inverse deformation correction image of the photovoltaic panel, matches the defect feature library, and obtains the photovoltaic panel defect matching and recognition results.

[0051] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A photovoltaic panel detection method based on machine vision, characterized in that: The following steps are involved: S1: Obtain the displacement data of multiple cells on the surface of the photovoltaic panel in the video, calculate the displacement vector gradient change rate of the photovoltaic cell connection point, divide the rigid area and the non-rigid area, and generate the photovoltaic panel area deformation classification result; S2: Based on the regional deformation classification result of the photovoltaic panel, the displacement change of the feature points between adjacent frames of the photovoltaic module encapsulation layer is calculated, the global deformation field of the photovoltaic panel is established, and the time series deformation trend matrix of the photovoltaic panel is generated; S3: Based on the photovoltaic panel time series deformation trend matrix, the overall displacement mean of the photovoltaic bracket fixing point is calculated, the deformation compensation parameters of the photovoltaic module encapsulation layer are adjusted, the deformation increment factor of the photovoltaic cell series connection area, the correction weight of the photovoltaic module non-rigid area and the deformation stability of the photovoltaic glass surface area are calculated, and the photovoltaic panel adaptive deformation compensation parameters are generated; S4: Based on the photovoltaic panel adaptive deformation compensation parameters, adjusting the positions of the characteristic points of the photovoltaic module welding points, calculating the mapping relationship, and generating a photovoltaic panel inverse deformation correction image; S5: Based on the photovoltaic panel inverse deformation correction image, the Euclidean distance of the morphological features between the abnormal area on the photovoltaic cell surface and the normal area on the photovoltaic cell surface is calculated to generate a photovoltaic panel defect matching recognition result.

2. The photovoltaic panel detection method based on machine vision according to claim 1, characterized in that: The photovoltaic panel regional deformation classification results include the displacement vector gradient change rate of the rigid area, non-rigid area, and photovoltaic cell connection point; the photovoltaic panel time series deformation trend matrix includes the displacement change of the feature points between adjacent frames of the photovoltaic module encapsulation layer and the global deformation field of the photovoltaic panel; the photovoltaic panel adaptive deformation compensation parameters include the overall displacement mean of the photovoltaic bracket fixing point, the deformation compensation parameters of the photovoltaic module encapsulation layer, the deformation increment factor of the photovoltaic cell series connection area, the correction weight of the photovoltaic module non-rigid area, and the deformation stability of the photovoltaic glass surface area; the photovoltaic panel inverse deformation correction image includes the position adjustment and mapping relationship of the feature points of the photovoltaic module welding point; the photovoltaic panel defect matching and identification results include the Euclidean distance of the morphological features of the abnormal area on the surface of the photovoltaic cell and the Euclidean distance of the morphological features of the normal area of ​​the photovoltaic cell.

3. The photovoltaic panel detection method based on machine vision according to claim 2, characterized in that: The steps for obtaining the photovoltaic panel regional deformation classification result are specifically as follows: S101: Obtain displacement data of the solar cells on the surface of the photovoltaic panel, calculate the local displacement vector, obtain the gradient of the connection point displacement vector, and obtain the gradient distribution of the displacement vector; S102: calling the displacement vector gradient distribution, calculating the gradient value, screening high and low gradient areas, determining the distribution density, dividing the rigid and non-rigid areas, and obtaining the regional rigidity classification matrix; S103: calling the regional rigid classification matrix, analyzing the non-rigid regional connection points, calculating the deformation center, and calculating the deformation classification index in combination with the rigid regional distribution, using the formula: ； Calculate the regional deformation classification value of the photovoltaic panel, filter the deformation category, and obtain the regional deformation classification result of the photovoltaic panel; in, Represents the regional deformation classification value of the photovoltaic panel, Representative The displacement vector gradient of the non-rigid region, represents the distance from the region to the adjacent rigid region, Adjust the parameters for the deformation. Represents the first The deformation strength of a point, represents the average deformation strength of the rigid region adjacent to the point, represents the total number of non-rigid regions, Represents the total number of adjacent rigid regions.

4. The photovoltaic panel detection method based on machine vision according to claim 3, characterized in that: The specific steps for obtaining the photovoltaic panel time series deformation trend matrix are as follows: S201: based on the photovoltaic panel regional deformation classification result, obtaining feature points between adjacent frames of the encapsulation layer, calculating the displacement change of the feature points, and obtaining a feature point displacement change matrix; S202: calling the feature point displacement change matrix, calculating the deformation vectors of multiple feature points in the global coordinate system, integrating local deformation information, and establishing a global deformation field of the photovoltaic panel; S203: calling the photovoltaic panel global deformation field, calculating the deformation trend values ​​of multiple time steps in the time series, using the formula: ； Calculate the time series deformation characteristics and establish the photovoltaic panel time series deformation trend matrix; in, Represents the total deformation index under the time series, Representative The influence coefficient of the next time step, Representative The displacement change in the time step is, Indicates The displacement deviation correction amount of the time step, Representative The sensitivity of the change in the time step, Smoothing parameters that represent deformation trends, Indicates The displacement normalization parameter, Represents the total number of time steps for the reference.

5. The photovoltaic panel detection method based on machine vision according to claim 4, characterized in that: The steps for obtaining the photovoltaic panel adaptive deformation compensation parameters are specifically as follows: S301: Based on the photovoltaic panel time series deformation trend matrix, extract the displacement data of the photovoltaic bracket fixing point in the entire time series, and calculate the overall displacement mean of the photovoltaic bracket; S302: calling the overall displacement mean of the photovoltaic support, calculating the parameters of the photovoltaic module encapsulation layer, adjusting the parameters according to the compensation requirements of multiple module areas, and obtaining the photovoltaic module deformation compensation parameters; S303: Calling the photovoltaic module deformation compensation parameter, using the formula: ； Calculate the deformation increment factor of the photovoltaic cell series area; in, Represents the deformation increment factor of the photovoltaic cell series connection area, Representative The deformation change value of the series-connected area, represents the deformation weight of the region, Adjust the parameters for the deformation. Representing the region The deformation strength of the connection point, represents the mean deformation of the connection point, represents the total number of serially connected regions, The number of connection points representing the region; S304: calling the deformation increment factor of the photovoltaic cell series area, calculating the correction weight of the non-rigid area of ​​the photovoltaic module and the deformation stability of the photovoltaic glass surface, and establishing the photovoltaic panel adaptive deformation compensation parameters.

6. The photovoltaic panel detection method based on machine vision according to claim 5, characterized in that: The steps for obtaining the photovoltaic panel inverse deformation correction image are specifically as follows: S401: adjusting the position of the characteristic point of the photovoltaic module welding point based on the photovoltaic panel adaptive deformation compensation parameter, performing position correction according to the deformation compensation parameter, and obtaining the photovoltaic module welding point adjustment matrix; S402: calling the photovoltaic module welding point adjustment matrix, calculating the mapping relationship of the feature points before and after the adjustment, and establishing a coordinate transformation relationship for all adjustment points, using the formula: ； Calculate the mapping relationship between multiple feature points, obtain the spatial mapping matrix after deformation compensation, and obtain the photovoltaic module mapping relationship matrix; in, Representative Adjust the feature points to The mapping relationship value of the target feature points, Representative The feature point is The position offset in this adjustment, Representative The target feature point is The amount of correction in this adjustment, To adjust the offset correction factor, represents the normalization parameter, Representative The first adjustment The position of feature points in the original coordinates, Represents the new position of the same feature point after adjustment, represents the mapping smoothing parameter, Representative The adjusted feature point is The coordinate value in the round transformation, Representative The coordinate values ​​of the target feature points in the same round of transformation, Represents the number of adjustments, Represents the total number of feature points, Represents the total number of transformation operations; S403: calling the photovoltaic assembly mapping relationship matrix, applying transformation parameters to adjust the image pixel coordinates, reconstructing the corrected pixel distribution, performing image interpolation to fill the pixel gaps, and generating an inverse deformation correction image of the photovoltaic panel.

7. The photovoltaic panel detection method based on machine vision according to claim 6, characterized in that: The steps for obtaining the photovoltaic panel defect matching and identification results are specifically as follows: S501: based on the photovoltaic panel inverse deformation correction image, obtaining morphological features of abnormal areas and normal areas on the surface of the photovoltaic cell, extracting key morphological parameters, and obtaining a photovoltaic cell area morphological feature set; S502: calling the photovoltaic cell area morphological feature set, calculating the Euclidean distance of the morphological features between the abnormal area and the normal area, and performing distance weighted operations on multiple morphological feature dimensions, using the formula: ； Calculate the Euclidean distance of morphological features, screen the feature matching degree of abnormal areas, and obtain the Euclidean distance matrix of photovoltaic panel morphological features; in, Represents abnormal area With normal area The Euclidean distance of the morphological features between Representative The weighting factor of the morphological features, Represents abnormal area In the The value of the morphological feature dimension, Represents the normal area The values ​​on the same feature dimension, represents the norm parameter of the morphological feature distance calculation, represents the feature matching adjustment parameter, Represents abnormal area In the The values ​​on the auxiliary feature dimensions, Represents the normal area The values ​​on the same auxiliary feature dimension, Represents the total number of key features, Represents the total number of auxiliary features; S503: calling the Euclidean distance matrix of the photovoltaic panel morphological features, matching the morphological feature relationship between the abnormal area and the normal area, and establishing a photovoltaic panel defect matching and recognition result.

8. A photovoltaic panel detection system based on machine vision, characterized in that: According to the photovoltaic panel detection method based on machine vision according to any one of claims 1 to 7, the system comprises: The displacement data calculation module obtains the displacement data of the solar cells on the surface of the photovoltaic panel, calls the connection point coordinate information, calculates the displacement vector gradient change rate, selects points below the threshold to divide the rigid area, selects points above the threshold to divide the non-rigid area, integrates the rigid and non-rigid area distribution, and obtains the photovoltaic panel area deformation classification result; The deformation trend analysis module calls the coordinates of the feature points of the adjacent frames of the encapsulation layer based on the deformation classification results of the photovoltaic panel area, calculates the displacement change, integrates the global deformation field, and obtains the photovoltaic panel time series deformation trend matrix; The deformation compensation calculation module calls the displacement of the bracket fixed point based on the photovoltaic panel time series deformation trend matrix, calculates the overall mean, adjusts the compensation parameters, calculates the deformation increment factor, and obtains the photovoltaic panel adaptive deformation compensation parameters; The inverse deformation correction module adjusts the position of the characteristic points of the welding points based on the adaptive deformation compensation parameters of the photovoltaic panel, calculates the mapping relationship, and obtains the inverse deformation correction image of the photovoltaic panel; The defect recognition module calculates the Euclidean distance between the abnormal area and the normal area based on the inverse deformation correction image of the photovoltaic panel, matches the defect feature library, and obtains the photovoltaic panel defect matching recognition result.

Images