Photovoltaic module defect detection method and system based on image recognition

Abstract

The invention relates to the technical field of image recognition, in particular to a photovoltaic module defect detection method and system based on image recognition, and the method comprises the following steps: obtaining a photovoltaic module image of a photovoltaic module to be detected, and calculating an intensity mean value and a texture entropy value for a predefined region unit in the photovoltaic module image, and combining the intensity mean value and the texture entropy value of each region into a feature vector. According to the method, the intensity mean value and the texture entropy value of the predefined region unit in the photovoltaic module image are sequentially extracted, and the region double-feature vector set is constructed, so that partition modeling based on fine-grained image features is realized; feature mapping point clouds are generated in combination with spatial neighborhood features, and abnormal point screening is performed based on the local density of the point clouds, so that the recognition precision of local structure differences is improved; in the time dimension, images of a photovoltaic module at multiple historical moments are obtained, intensity mean values and texture entropy values are extracted one by one for areas corresponding to potential abnormal feature point sets, and a multi-time-point feature value column is established.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, and in particular to a photovoltaic module defect detection method and system based on image recognition. BACKGROUND

[0002] The technical field of image recognition is an important branch of artificial intelligence and computer vision, which is committed to automatically recognizing, analyzing and understanding the content of targets, scenes and features in images through computers.

[0003] The existing technology of image recognition mainly takes the overall image features or local significant features as the main analysis basis, lacks the recognition mechanism for implicit changes such as low-intensity changes, texture degradation and local sparse abnormalities, and thus it is difficult to discover some potential defects in time. For example, for the weak changes in texture of photovoltaic modules caused by material aging, it is difficult to identify abnormal features from a single image due to the unobvious intensity changes, which delays the fault response time. Therefore, improvement is needed. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art and to provide a photovoltaic module defect detection method and system based on image recognition.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme, a photovoltaic module defect detection method based on image recognition, comprising the following steps:

[0006] Obtain the photovoltaic module image of the photovoltaic module to be detected, calculate the intensity mean and texture entropy value of the predefined area unit in the photovoltaic module image, and combine the intensity mean and texture entropy value of each area into a feature vector to establish a region dual feature vector set;

[0007] Based on the region dual feature vector set, generate a feature mapping point cloud according to the spatial neighborhood features, judge the point cloud distribution density based on the feature mapping point cloud, judge the abnormal points, and obtain a potential abnormal feature point set;

[0008] Obtain the photovoltaic module images of the same photovoltaic module at different time points, extract the intensity mean and texture entropy value from each time point photovoltaic module image for the region corresponding to the potential abnormal feature point set, construct a multi-time point feature value column, calculate the statistical difference of the feature value between the continuous time points based on the multi-time point feature value column, and obtain a feature difference evolution sequence;

[0009] Based on the feature difference evolution sequence, analyze the rate and direction of the change of the feature difference measure of each region with time, establish a region evolution rate parameter, and based on the region evolution rate parameter, calculate the feature difference measure of the future period and generate a photovoltaic module defect warning judgment.

[0010] Preferably, the step of obtaining the region dual feature vector set is:

[0011] An image of the photovoltaic module to be detected is obtained, a plurality of predefined region units are marked in the image of the photovoltaic module, a pixel set of each predefined region unit is extracted respectively, and a region pixel set is obtained;

[0012] Based on the region pixel set, the average value of the gray value of all pixels in each predefined region unit is calculated to generate the intensity average value of the corresponding region unit, and the entropy value of the gray value distribution of each region pixel set is calculated to form the texture entropy value of the corresponding region unit;

[0013] Based on the intensity average value of the corresponding region unit and the texture entropy value of the corresponding region unit, a region dual feature vector set is formed.

[0014] Preferably, the step of obtaining the feature mapping point cloud is:

[0015] Based on the region dual feature vector set, the intensity average value and the texture entropy value of each region unit in the current image are extracted, and are numbered in order according to the region, the spatial coordinates of each region unit are identified, then the image data of the same spatial position of the photovoltaic module in the previous three days is retrieved, the intensity average value and the texture entropy value of the region unit with the same number in the previous three days are extracted, and a three-time-point historical feature sequence group is generated;

[0016] According to the three-time-point historical feature sequence group, the spatial neighborhood features of each region unit in the current image are combined to calculate a structure disturbance index;

[0017] Based on the structure disturbance index, the structure disturbance indexes of all region units in the current image are sorted in ascending order according to the size, and the region units with the structure disturbance indexes in the top 25% of all region units are selected, the feature vectors of the region units are combined into a target mapping vector set, and a feature mapping point cloud is generated.

[0018] Preferably, the step of obtaining the potential abnormal feature point set is:

[0019] Based on the feature mapping point cloud, the two-dimensional coordinate values of each region unit in the point cloud are read, a fixed radius is set as a search range, all point cloud coordinates are traversed, the number of point pairs forming a geometric neighborhood relationship within the fixed radius is identified, the number of neighboring points of each point is obtained, and a neighborhood coordinate set is constructed;

[0020] According to the neighborhood coordinate set and the number of neighboring points, a density outlier index of each point is calculated;

[0021] Based on the density outlier index, all region units are sorted in descending order according to the density outlier index, and region units with a density outlier index higher than the median of the density outlier index of all region units are screened to generate a set of potential abnormal feature points.

[0022] Preferably, the step of obtaining the multi-time point feature value column is:

[0023] Obtain photovoltaic module images formed by shooting the same photovoltaic module at different collection times, analyze the time labels of the images, extract pixel data of the corresponding region unit from each photovoltaic module image according to the region unit index information recorded by the set of potential abnormal feature points, and form a multi-time pixel set of the corresponding region unit;

[0024] Based on the multi-time pixel set of the corresponding region unit, calculate the average value of pixel gray value and the texture entropy value of the gray distribution of each region unit, record and pair the region unit feature parameter combination in sequence, arrange the region unit feature parameter combination according to the time label sequence, and generate a multi-time feature parameter combination sequence of the region unit.

[0025] Based on the multi-time feature parameter combination sequence of the region unit, all multi-time feature parameter combination sequences of the region units are uniformly summarized to generate a multi-time point feature value column.

[0026] Preferably, the step of obtaining the feature difference evolution sequence is:

[0027] Based on the multi-time point feature value column, the intensity mean value and the texture entropy value of each region unit at each time point are extracted according to the region unit index, the feature parameters of adjacent time points are paired one by one, the difference value between the intensity mean value and the texture entropy value of adjacent time points is calculated, and the adjacent time point feature difference value sequence of each region unit is obtained.

[0028] According to the adjacent time point feature difference value sequence, the absolute value of the difference value between the intensity mean value and the texture entropy value of each region unit at consecutive time points is calculated, the sum of the absolute values of the intensity mean value and the texture entropy value difference of the same region unit is calculated, and then divided by the number of corresponding consecutive time point feature parameters to calculate the average difference value of the consecutive time point feature parameters, and the statistical difference of the region unit at consecutive time points is obtained.

[0029] Based on the statistical difference of the region unit at consecutive time points, the statistical difference value is recorded one by one according to the region unit index sequence, and the time sequence data sequence recording the change of the statistical difference of the feature parameters of all region units at consecutive time points is established with the region unit as the index key, and the feature difference evolution sequence is generated.

[0030] Preferably, the step of obtaining the region evolution rate parameter is:

[0031] Based on the feature difference evolution sequence, the statistical difference of each regional unit at continuous time points is extracted one by one based on the regional unit index, the numerical difference between the statistical values of the feature difference of each regional unit at adjacent time labels is calculated according to the time label, and the initial rate of change of the difference of each regional unit is determined by dividing the time interval, and an initial rate value set of the feature difference is generated.

[0032] According to the initial rate value set of the feature difference, the positive and negative signs of the initial rate value of the feature difference of each regional unit are judged respectively, if the rate value is positive, it is marked that the feature difference of the regional unit shows an increasing trend, if the rate value is negative, it is marked that the feature difference of the regional unit shows a decreasing trend, and a regional unit difference evolution trend set is obtained.

[0033] Based on the regional unit difference evolution trend set, the number of times of occurrence of the increasing trend and the decreasing trend of the feature difference of each regional unit in all continuous time periods is counted, and the number of times of occurrence is multiplied by the average absolute value of the rate of the corresponding regional unit in the initial rate value set of the feature difference to form a regional evolution rate parameter.

[0034] Preferably, the acquisition step of the photovoltaic module defect early warning judgment is:

[0035] Based on the regional evolution rate parameter, the evolution rate and the corresponding time interval number are extracted one by one according to the regional unit index, the statistical value of the feature difference at the nearest time is taken as the current starting value, the evolution rate value of each regional unit is multiplied by the time interval number of the prediction step, and is added to the current starting value to generate a feature difference degree prediction value set of each regional unit in the future time period.

[0036] According to the feature difference degree prediction value set of each regional unit in the future time period, the prediction value of each regional unit is compared with the maximum value, the average value and the upper threshold of the stable interval of the feature difference degree in the historical time period in sequence, if the prediction value exceeds the maximum value, the average value or the upper threshold of the stable interval, it is marked as having a future feature abnormal trend, otherwise it is marked as being within the acceptable range of feature change, and a regional unit feature difference out-of-limit judgment label is obtained.

[0037] Based on the regional unit feature difference out-of-limit judgment label, all regional unit indexes with future feature abnormal trend are merged, and a defect identification table is generated combining the photovoltaic module number and the regional unit index to generate a photovoltaic module defect early warning judgment.

[0038] The application also provides a photovoltaic module defect detection system, which comprises:

[0039] An image feature extraction module acquires a photovoltaic module image of a photovoltaic module to be detected, calculates intensity mean values and texture entropy values for predefined region units in the photovoltaic module image, combines the intensity mean values and the texture entropy values of each region into a feature vector, and establishes a region dual feature vector set;

[0040] A feature point cloud generation and anomaly detection module generates a feature mapping point cloud according to spatial neighborhood features based on the region dual feature vector set, judges point cloud distribution density based on the feature mapping point cloud, judges abnormal points, and acquires a potential abnormal feature point set;

[0041] A time sequence feature difference analysis module acquires photovoltaic module images of a same photovoltaic module at different time points, extracts intensity mean values and texture entropy values from each time point photovoltaic module image for a region corresponding to the potential abnormal feature point set, constructs a multi-time point feature value column, calculates a statistical difference between feature values at consecutive time points based on the multi-time point feature value column, and obtains a feature difference evolution sequence;

[0042] A defect evolution trend prediction module analyzes a rate and a direction of change of a region feature difference metric with time based on the feature difference evolution sequence, establishes a region evolution rate parameter, calculates a feature difference metric in a future time period based on the region evolution rate parameter, and generates a photovoltaic module defect early warning judgment.

[0043] Compared with the prior art, the present application has the following advantages and positive effects:

[0044] The present application realizes partition modeling based on fine-grained image features by sequentially extracting intensity mean values and texture entropy values of predefined region units in a photovoltaic module image and constructing a region dual feature vector set, improves recognition accuracy of local structural differences by generating a feature mapping point cloud in combination with spatial neighborhood features and screening abnormal points based on local density of the point cloud, realizes characterization of a local region change trend by acquiring images of a photovoltaic module at multiple historical time points, extracting intensity mean values and texture entropy values for regions corresponding to a potential abnormal feature point set one by one, establishing a multi-time point feature value column, and quantifying a feature evolution process by a statistical difference between consecutive time points, and generates a defect early warning judgment in combination with a historical evolution law by further deriving a feature evolution rate and a direction of each region from a continuous difference value sequence and extending the calculation result to a future time period. Thus, by introducing time sequence evolution analysis on the basis of spatial feature extraction and in combination with partition tracking, difference trend analysis, and future prediction calculation, the present application breaks through the problem of insufficient capture of potential structural degradation in traditional static image recognition, realizes dynamic recognition and early warning of local potential change features of a photovoltaic module, and improves timeliness and accuracy of fault response. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1A schematic diagram of the steps of the present application. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0047] Please refer to Figure 1 The present application provides a technical solution, a photovoltaic module defect detection method based on image recognition, comprising the following steps:

[0048] Obtain the photovoltaic module image of the photovoltaic module to be detected, calculate the intensity mean value and the texture entropy value of the predefined area unit in the photovoltaic module image, and combine the intensity mean value and the texture entropy value of each area into a feature vector to establish a region dual feature vector set;

[0049] Based on the region dual feature vector set, generate a feature mapping point cloud according to the spatial neighborhood features, judge the point cloud distribution density based on the feature mapping point cloud, judge the abnormal points, and obtain a potential abnormal feature point set;

[0050] Obtain the photovoltaic module images of the same photovoltaic module at different time points, extract the intensity mean value and the texture entropy value from each time point photovoltaic module image for the area corresponding to the potential abnormal feature point set, construct a multi-time point feature value column, calculate the statistical difference of the feature value between the continuous time points based on the multi-time point feature value column, and obtain a feature difference evolution sequence;

[0051] Based on the feature difference evolution sequence, analyze the rate and direction of the change of the feature difference metric of each area with time, establish an area evolution rate parameter, and calculate the feature difference metric of the future period based on the area evolution rate parameter to generate a photovoltaic module defect early warning judgment.

[0052] The acquisition step of the region dual feature vector set is:

[0053] Obtain the photovoltaic module image of the photovoltaic module to be detected, mark a plurality of predefined area units in the photovoltaic module image, respectively extract the pixel set of each predefined area unit, and obtain the region pixel set;

[0054] Based on the region pixel set, average the gray value of all pixels in each predefined area unit to generate the intensity mean value of the corresponding area unit, and calculate the entropy value of the gray value distribution of each region pixel set to form the texture entropy value of the corresponding area unit;

[0055] Based on the intensity mean value of the corresponding area unit and the texture entropy value of the corresponding area unit, the region dual feature vector set is formed.

[0056] Specifically, after obtaining the photovoltaic module image of the photovoltaic module to be detected, first, according to the size information of the image, for example, an image with a resolution of 1024x768 pixels, the image is divided into a 16x12 grid, so as to mark 192 predefined area units with a size of 64x64 pixels in the photovoltaic module image, and each predefined area unit is assigned a unique spatial index number in the order from left to right and from top to bottom, for example, the index of the area unit in the first row and the first column is (0, 0), the index of the area unit in the first row and the second column is (0, 1), and so on. Then, the system will traverse all 192 predefined area units. For each area unit, the system reads the gray values of all 4096 pixels in the area one by one according to the coordinate range of the area unit in the image, and stores these gray values as a set, thereby obtaining a region pixel set corresponding to each predefined area unit.

[0057] Based on the region pixel set obtained in the previous process, the system will perform double-feature calculation for each predefined area unit. First, for the calculation of the intensity mean value, the system reads all the pixel gray values in a region pixel set, adds all the values, and then divides the sum by the total number of pixels in the region (for example, the total number of pixels in a 64x64 region is 4096), thereby generating the intensity mean value of the corresponding area unit. Second, for the calculation of the texture entropy value, the system first counts the number of pixels of each gray level (0 to 255) in the region pixel set to construct a 256-item gray histogram, then divides the count value of each item by the total number of pixels in the region to obtain the occurrence probability of each gray level, and finally calculates the entropy value of the gray value distribution of the region pixel set according to the Shannon entropy calculation formula to form the texture entropy value of the corresponding area unit. The calculation formula is: where H represents the texture entropy value of the region, L is the total number of gray levels, usually taking a value of 256, and p(i) is the probability of the pixel with a gray value of i occurring in the region.

[0058] Based on the intensity mean value of each corresponding region unit and the texture entropy value of each corresponding region unit calculated in the preceding step, the system will perform a summary operation. Specifically, the system will create a two-dimensional feature vector for each predefined region unit, with the first dimension storing the intensity mean value of the region and the second dimension storing the texture entropy value of the region. For example, for the region unit with index (0, 0), if its calculated intensity mean value is 128.5 and its texture entropy value is 4.2, its corresponding feature vector is (128.5, 4.2). The system will generate such two-dimensional feature vectors for all predefined region units (e.g., 192) in the image in sequence, and associate all these two-dimensional feature vectors with their corresponding region spatial index numbers. Finally, the system will organize all the region unit indices and their corresponding two-dimensional feature vectors into a set, which is the region dual feature vector set.

[0059] The feature mapping point cloud acquisition step is:

[0060] Based on the region dual feature vector set, the intensity mean value and texture entropy value of each region unit in the current image are extracted, and the spatial coordinates of each region unit are identified in sequence. Then, the image data of the same spatial position of the photovoltaic module in the past three days is retrieved, the intensity mean value and texture entropy value of the same numbered region unit in the past three days are extracted, and a three-time-point historical feature sequence group is generated.

[0061] According to the three-time-point historical feature sequence group, the spatial neighborhood features of each region unit in the current image are combined to calculate the structural disturbance index, and the calculation formula is:

[0062]

[0063] ZD m is the structural disturbance index of the mth region unit, a m1 , a m2 are the intensity mean value and texture entropy value of the mth region unit in the current image, a j1 , a j2 are the intensity mean value and texture entropy value of the corresponding region of the region unit in the image of the past jth day, u m is the horizontal position coordinate value of the mth region unit in the current image, u k is the horizontal position coordinate value of the kth spatial neighborhood unit of the region unit, N k is the number of spatial neighborhood units of the region unit, and ∈ is a normal number to prevent division by zero.

[0064] Based on the structural disturbance index, the structural disturbance indexes of all region units in the current image are sorted in ascending order according to the size, and region units with the structural disturbance indexes in the top 25% of all region units are selected. The feature vectors of the region units are combined into a target mapping vector set to generate a feature mapping point cloud.

[0065] Specifically, based on the region double feature vector set, the system first traverses each entry in the set, extracts the intensity mean and texture entropy value of each predefined region unit in the current image, and assigns a unique ordinal number starting from 0 to each region unit according to the row and column order set when the grid is previously divided, while recording its two-dimensional spatial coordinates in the image grid. For example, a region unit located in the 3rd row and 5th column has a number of 36 and a spatial coordinate of (3, 5). Subsequently, the system initiates a retrieval request to the image data repository according to the unique identifier of the current photovoltaic module, to find the photovoltaic module images taken in the same collection period of the past three consecutive natural days (i.e. the day before, the day before, and the day before) of the module. After successful retrieval, the system accurately locates each numbered region unit in the corresponding physical location in the historical images for the three historical images using the recorded region unit spatial coordinates, and re-executes the intensity mean and texture entropy value calculation for the pixel data at that location. The double feature values calculated daily are bound to the corresponding region unit number and date label. Finally, all region unit feature data at the three historical time points are integrated to form a structured data set with region unit number as index, each index containing three historical double feature vectors, i.e. a three-time-point historical feature sequence group.

[0066]

[0067] The formula has the benefit that it combines the amount of change of the region feature in the time dimension with the structural difference in the spatial dimension, achieving accurate quantitative evaluation of the potential defect region of the photovoltaic module. The first part of the formula, the Euclidean distance term measures the degree of evolution of the characteristics (intensity and texture) of a single region unit over time, and can capture local visual feature changes caused by aging, stains or internal damage. The second part of the formula, the spatial neighborhood disturbance term evaluates the isolation or particularity of the target region unit in the spatial structure by calculating the relative difference in position between the target region unit and its neighborhood units. Multiplying the two parts makes the structural disturbance index significantly higher only when a region unit has both significant changes in time and inconsistency with its surroundings in space. This design can effectively filter out global and uniform changes caused by changes in environmental light, overall uniform dust accumulation, etc., thereby locking regions with locally abnormal evolution characteristics.

[0068] am1 with a m2 The steps for obtaining the parameters are as follows: these two parameters represent the intensity mean and texture entropy value of the mth regional unit in the current image to be detected, respectively. They are calculated based on the current photovoltaic module image through the above steps and do not require additional settings. The system directly reads from the generated regional dual feature vector set according to the regional unit index m. For example, for the regional unit with index 50, the system finds its intensity mean as 135.6 and texture entropy as 5.1 from the regional dual feature vector set, then a m1 =135.6, a m2 =5.1.

[0069] a j1 with a j2 The steps for obtaining the parameters are as follows: these two parameters represent the intensity mean and texture entropy value of the corresponding area of ​​the mth regional unit in the historical image of the past jth day (j=1, 2, 3 represent the previous day, previous two days, and previous three days respectively). The values ​​are derived from the three-point historical feature sequence group generated in the previous step. The system extracts from the sequence group according to the currently processed regional unit index m and the historical day index j. For example, for the regional unit with an index of 50, the eigenvalue of the previous day (j=1) is (138.2, 5.0), the eigenvalue of the previous two days (j=2) is (139.1, 4.9), and the eigenvalue of the previous three days (j=3) is (138.8, 4.9). These values ​​are then used in sequence during the calculation.

[0070] u m with u k The steps of obtaining the parameters are as follows: the two parameters represent the horizontal position coordinate values ​​of the mth region unit and its kth spatial neighboring unit in the current image, respectively. The coordinate values ​​are determined when the image is divided into a grid and represent the column index of the region unit in the grid. For example, if the image is divided into 16 columns, the horizontal position coordinate value ranges from 0 to 15. For a region unit located in the 8th row and the 6th column (index m = 117), its horizontal position coordinate u m =5, its spatial neighborhood is defined as the 8 area units around it, and the horizontal position coordinates of these neighborhood units are u k They are 4, 5, 6, 4, 6, 4, 5, 6 respectively.

[0071] N k The steps to obtain the parameter are as follows: this parameter represents the number of spatial neighboring units of the area unit, and its value depends on the position of the area unit in the image grid. For an area unit located inside the image and not in contact with the boundary, there are 8 neighboring units around it, so N k =8, for the area unit located at the edge of the image (not the corner), it has 5 neighboring units, N k=5, for the area unit located in the corner of the image, it has 3 neighboring units, N k =3.

[0072] The steps to obtain the ∈ parameter are as follows: the parameter is a minimal positive constant introduced to prevent division by zero errors. For example, set ∈ = 1 × 10 -6 .

[0073] Calculation process:

[0074] The structural perturbation index ZD of the regional unit with the calculation index m=117 is 117 For example, the region unit is located inside the image and does not touch the boundary.

[0075] Get the parameter values ​​as follows:

[0076] Current eigenvalue: a m1 =135.6, a m2 =5.1;

[0077] Historical characteristic values:

[0078] j=1:a 11 =138.2, a 12 =5.0;

[0079] j=2:a 21 =139.1, a 22 =4.9;

[0080] j=3:a 31 =138.8, a 32 =4.9;

[0081] Position coordinates: u m =5, the horizontal position coordinates u of its 8 neighboring units k They are {4, 5, 6, 4, 6, 4, 5, 6} respectively, and the number of neighbors is N k =8, anti-zero constant∈=1×10 -6 .

[0082] First, calculate the spatial neighborhood disturbance term (denoted as S m ):

[0083] 0.1375;

[0085] Then, the product of the time evolution term and the spatial neighborhood disturbance term of the three historical points is calculated respectively:

[0086] For j = 1:

[0087]

[0088] For j = 2:

[0089]

[0090] For j = 3:

[0091]

[0092] Finally, the structure disturbance index ZD is calculated 117 :

[0093]

[0094] The result shows that the structure disturbance index of the 117th regional unit is 0.4269, which comprehensively reflects the degree of feature change of the region in the past three days and its spatial structure difference with the surrounding regions. The value itself is a relative measure, which needs to be compared in the calculation results of all regional units. A higher value (for example, ranked first among all regional units) means that the regional unit shows stronger spatio-temporal instability compared to other regions, and has a higher possibility of being a potential defect point.

[0095] Based on the calculated structure disturbance indexes of all regional units, the system pairs these index values with their respective regional unit indexes to form a tuple list containing (region index, structure disturbance index). Then, all tuples in the list are sorted in ascending order according to the structure disturbance index values to obtain an ordered list from low to high. Next, to screen out the most disturbed regions, the system uses the percentile screening method. Here, the "top 25%" specifically refers to the top quarter of the regional units in terms of structure disturbance index values. Therefore, the system calculates the 75th percentile of the total number of regions as a threshold. For example, for a total of 192 regional units, the percentile index is 192 x (1-0.25) = 144. The system will select the 48 regional units from the 145th to the 192nd in the sorted list. These selected regional units are considered high-risk due to their highest structure disturbance indexes. Finally, the system returns to the originally generated regional dual feature vector set according to the indexes of these screened high-risk regional units, extracts their original intensity mean and texture entropy values in the current image, and collects these two-dimensional feature vectors to form a set of 48 two-dimensional points, which is the feature mapping point cloud.

[0096] The steps to obtain the potential abnormal feature point set are:

[0097] Based on the feature mapping point cloud, the two-dimensional coordinate values of each regional unit in the point cloud are read, a fixed radius is set as a search range, all point cloud coordinate pairs are traversed, the number of point pairs forming a geometric neighborhood relationship within the fixed radius is identified, the number of neighboring points of each point is obtained, and a neighborhood coordinate set is constructed;

[0098] According to the neighborhood coordinate set and the number of neighboring points, the density outlier index of each point is calculated, and the calculation formula is:

[0099] R q = ln(1+D avg (q));

[0100]

[0101] wherein, R q is the density outlier index of the qth regional unit in the feature mapping point cloud, D avg (q) is the neighborhood average distance of the point within the search radius r0, (x q ,y q ) is the two-dimensional coordinate value of the qth regional unit, (x s ,y s ) is the two-dimensional coordinate value of the adjacent regional unit, h q is the number of adjacent points falling within the radius r0, and r0 is a specified fixed radius constant;

[0102] Based on the density outlier index, all regional units are sorted in descending order according to the density outlier index, and regional units with a density outlier index higher than the median of the outlier index of all regional units are screened to generate a set of potential abnormal feature points.

[0103] Specifically, based on the feature mapping point cloud, the system first reads the two-dimensional coordinate value composed of the intensity mean and the texture entropy value of each point in the point cloud, that is, each high-risk area unit. Subsequently, in order to quantify the isolation degree of each point in the feature space, a fixed radius needs to be set. The determination process of the radius is as follows: the system first performs k-neighbor analysis on all points in the feature mapping point cloud. Specifically, the distance from each point to its fourth nearest neighbor point is calculated, and these distance values are sorted in ascending order. A k-distance graph is drawn. By observing the "inflection point" of the graph, that is, the point with the most significant change in slope, an optimal neighborhood radius is determined. The distance value corresponding to the inflection point reflects the transition of the data points from the dense area to the sparse area, which can effectively distinguish core points and noise points. For example, after calculation and drawing analysis, it is determined that the distance corresponding to the inflection point is 5.0. Therefore, the fixed radius r0 is set to 5.0. After setting the radius, the system will traverse each point in the point cloud as a center point, and then traverse all other points in the point cloud. The Euclidean distance between the center point and each other point is calculated. If the distance is less than or equal to 5.0, the other point is identified as the geometric neighborhood point of the center point, and the number of neighboring points of the center point is accumulated. At the same time, its coordinates are recorded in the neighborhood coordinate set of the center point. After completing the traversal of all points, the number of neighboring points corresponding to each point and a neighborhood coordinate set containing all neighborhood point coordinates of each point are obtained.

[0104] Formula: R q = ln(1 + D avg (q)), The formula has the benefit of constructing a robust, local density-based outlier measurement index for identifying abnormal points in the feature space. The core of the formula is D avg (q), which evaluates the local sparsity of a point by calculating the average distance from the point to all points in its neighborhood. The farther a point is from its neighbors, the larger its D avg (q) value, indicating that it is more likely to be an abnormal point. In particular, the formula distinguishes between points with neighbors and completely isolated points (h q = 0) through a piecewise function. The average distance of isolated points is directly set to the search radius r0, which gives isolated points a certain, higher penalty value, ensuring that the most isolated points can obtain the highest outlier score. Finally, by taking the natural logarithm (ln) of 1 + D avg (q), the average distance value, which can have a wide range, can be mapped to a smoother, more easily comparable scale, suppressing the influence of extreme distance values on the overall ranking.

[0105] (x q , y qThe acquisition step of the (x, y) parameter is to obtain the two-dimensional coordinate values of the qth region unit currently being evaluated in the feature mapping point cloud, which correspond to the intensity mean value and the texture entropy value of the region unit respectively. The values are obtained by calculating the structure disturbance index and screening in the previous step. When calculating the density outlier index, the values can be directly read from the feature mapping point cloud dataset according to the index q. For example, if a point with coordinates (195.2, 1.8) is selected from the feature mapping point cloud for calculation, then x q = 195.2 and y q = 1.8.

[0106] The acquisition step of the (x s , y s ) parameter is to obtain the two-dimensional coordinate values of the s-th neighboring point of point q. The neighboring points are determined by searching with point q as the center and a fixed radius r0 in the previous step, and their coordinate value set is stored in the neighborhood coordinate set constructed for point q. When calculating, the system will traverse all coordinate points in the neighborhood coordinate set. For example, if the neighborhood coordinate set of point q contains two points with coordinates (192.1, 2.0) and (198.0, 1.5), then (x s , y s ) will take these two values in turn in the summation calculation.

[0107] The acquisition step of the h q parameter is to obtain the number of neighboring points falling within the circular region centered at point q with a radius of r0. This value is obtained by counting during the construction of the neighborhood coordinate set and is stored in association with each point q as the "number of neighboring points". The system can directly read this value when calculating. The value reflects the local density around point q. For example, for the aforementioned point q, its neighborhood coordinate set contains two points, so the number of neighboring points h q = 2.

[0108] The acquisition step of the r0 parameter is to obtain a fixed radius constant used to define the neighborhood range. The value is determined by analyzing the global structural characteristics of the entire feature mapping point cloud. The specific method is as follows: for each point in the point cloud, calculate the distance to all other points and find the k-th nearest distance (k-distance). Typically, k is 4. Arrange all k-distance values in ascending order and plot a k-distance graph. The distance value corresponding to the "inflection point" where the curve slope increases sharply in the graph is the ideal r0, as it marks the natural transition boundary from the core object (dense region) to the outlier object (sparse region). Through this analysis of a feature mapping point cloud containing 48 data points, it is determined that the inflection point appears at a distance value of 5.0, so r0 is set to 5.0.

[0109] Calculation process:

[0110] To calculate the density outlier index R of a point q in the computed feature-mapped point cloud q Take the example.

[0111] The parameter values are obtained as follows:

[0112] The coordinates of the point to be calculated: (x q ,y q ) = (195.2, 1.8);

[0113] According to the previous steps, the point has two neighboring points in the neighborhood with a radius r0 = 5.0, i.e. h q = 2.

[0114] The coordinates of the two neighboring points are respectively: (x s1 ,y s1 ) = (192.1, 2.0), (x s2 ,y s2 ) = (198.0, 1.5).

[0115] Since h q = 2 > 0, first calculate the neighborhood average distance D avg (q):

[0116]

[0117] Calculate the distance to the first neighboring point:

[0118]

[0119] Calculate the distance to the second neighboring point:

[0120]

[0121] Calculate the average distance:

[0122]

[0123] Next, calculate the density outlier index R q :

[0124] R q = ln(1 + D avg (q)) = ln(1 + 2.961) = ln(3.961) ≈ 1.376;

[0125] The result shows that the density outlier index of point q is 1.376, which quantifies the degree of isolation of the point in its local neighborhood. The higher the index value, the sparser the distribution of the point in its feature space, and the greater the difference from the surrounding points, so the higher the likelihood of being considered an outlier.

[0126] Based on the calculated density outlier index of each area unit, the system integrates these index values with the corresponding area unit index to form a list containing all high-risk area units and their outlier indexes. Subsequently, the system sorts this list in descending order of density outlier index values, so that the area unit with the highest degree of abnormality is ranked first. Next, to further filter out the most abnormal candidates from this batch of high-risk points, the system uses the median as the screening threshold. The median is the value located in the middle position after sorting all outlier index values. The median is chosen as the threshold because it is not sensitive to extreme values in the data and has strong robustness, which can objectively divide the data into two parts with high and low degrees of abnormality. The specific operation is as follows: the system calculates the median of all density outlier indexes in the ordered list. For example, if there are 48 outlier index values, the median is the average of the 24th and 25th values after sorting. After obtaining the median threshold, the system traverses the sorted list. All area units with a density outlier index higher than the median threshold are finally confirmed as abnormal, and their indexes are extracted to form a set of potential abnormal feature points.

[0127] The acquisition step of the multi-time point feature value column is:

[0128] The image of the photovoltaic module formed by shooting at different collection times is obtained, the time label of the image is analyzed, the pixel data of the corresponding area unit is extracted from each photovoltaic module image according to the area unit index information recorded by the potential abnormal feature point set, and a multi-time pixel set of the corresponding area unit is formed.

[0129] Based on the multi-time pixel set of the corresponding area unit, the average value of the pixel gray value and the texture entropy value of the gray distribution are calculated for each area unit, and the area unit feature parameter combinations are recorded and paired in turn, and the area unit feature parameter combinations are arranged in order according to the time label to generate a multi-time feature parameter combination sequence of the area unit.

[0130] Based on the multi-time feature parameter combination sequence of the area unit, all multi-time feature parameter combination sequences of the area units are uniformly summarized to generate a multi-time point feature value column.

[0131] Specifically, after obtaining the photovoltaic module image sequence formed by shooting the specified photovoltaic module at different collection times (for example, 12 o'clock every day for 10 consecutive days), the system first parses the accurate time label from the metadata of each image, for example, "2023-10-26-12-00-05", and sorts the image sequence strictly in time sequence according to these time labels. Then, the system calls the potential abnormal feature point set generated in the previous step, which contains the index list of the region units preliminarily judged as abnormal, for example, {50, 117, 189}. The system will traverse each index in this list. For each index (such as index 50), the system will traverse the entire time sequence image sequence again. In each image, the pixel coordinate range corresponding to index 50 is calculated (for example, if the image is 1024x768 and the region unit is 64x64, index 50 corresponds to the 4th row and the 3rd column, and its pixel coordinate range is x from 128 to 191 and y from 192 to 255), and all pixel gray value data in this rectangular region is extracted. The pixel data is associated and stored with the corresponding time label and region index. After completing the extraction of pixel data of all potential abnormal regions at all times, a region unit index-based key and a value containing multiple (time label, pixel data) pairs are finally formed, which is the corresponding region unit multi-time pixel set.

[0132] Based on the corresponding region unit multi-time pixel set, the system starts a nested processing flow. The outer loop traverses each potential abnormal region unit index, and the inner loop traverses the pixel set of the region unit at all different times. In each inner loop, for example, when processing the pixel data of the region unit with index 50 at "2023-10-26-12-00-05", the system first sums all pixel gray values in the pixel set and divides the total number of pixels to calculate the intensity mean value of the region at that time. At the same time, the system will count the number of pixels of each gray level (0 to 255) in the pixel set to construct a gray histogram, and calculate the Shannon entropy of the gray value distribution based on the histogram as the texture entropy value of the region at that time. After the calculation is completed, the obtained intensity mean value and texture entropy value are paired into a two-dimensional region unit feature parameter combination, for example, (145.2, 4.8), and are recorded together with the corresponding time label and region index. When the pixel data of all time points of a region unit is processed, the system arranges these region unit feature parameter combinations generated in time sequence to generate a complete time sequence for the region unit, that is, the region unit multi-time feature parameter combination sequence.

[0133] Based on the region unit multi-time feature parameter combination sequence generated for each potential abnormal region unit, the system performs a final data aggregation and formatting operation to integrate these independent, region-based time series data into a unified, flattened data structure. Specifically, the system creates a structured data list, where each row represents a specific region's feature observation value at a specific time. This list contains at least four columns: time label, region unit index, intensity mean, and texture entropy. The system parses each region unit's multi-time feature parameter combination sequence one by one, converting each (time label, feature parameter combination) entry into a row of data in the new list. For example, for region 50 at "2023-10-26-12-00-05" with features (145.2, 4.8), it is recorded as a row ["2023-10-26-12-00-05", 50, 145.2, 4.8]. This process covers all potential abnormal region units at all observation time points, resulting in a comprehensive list of feature values for all related regions and time points, which is the multi-time point feature value column.

[0134] The feature difference evolution sequence acquisition step is:

[0135] Based on the multi-time point feature value column, the intensity mean and texture entropy values of each region unit at each time point are extracted according to the region unit index. The feature parameters of adjacent time points are paired one by one, and the difference values of intensity mean and texture entropy between adjacent time points are calculated to obtain the adjacent time point feature difference value sequence of each region unit.

[0136] According to the adjacent time point feature difference value sequence, the absolute values of the intensity mean and texture entropy difference values between consecutive time points for each region unit are calculated. The sum of the absolute values of the intensity mean and texture entropy difference values for the same region unit is calculated, and then divided by the number of corresponding consecutive time point feature parameters to calculate the average difference value of consecutive time point feature parameters, obtaining the statistical difference of the region unit consecutive time points.

[0137] Based on the statistical difference of the region unit consecutive time points, the statistical difference values are recorded one by one according to the region unit index order, and a time sequence data sequence recording the feature parameter statistical difference changes between consecutive time points for all region units is established, generating the feature difference evolution sequence.

[0138] Specifically, based on the multi-time point feature value column, the system first groups the data according to the region unit index. Taking the region unit index 50 as an example, the system will filter out all records with index 50 from the multi-time point feature value column, and arrange them in ascending order according to the time label to form a time series containing the intensity mean value and texture entropy value of the region in the next 10 days, for example, the first day feature is (145.2, 4.8), the second day is (142.7, 5.0), the third day is (140.1, 5.1), and so on. Then, the system performs sliding window processing on this time series, and the window size is 2. The feature parameters of adjacent time points (for example, the first day and the second day, the second day and the third day) are paired one by one. For each pair of adjacent time point feature parameters, the difference value of the intensity mean value and the difference value of the texture entropy value are calculated independently. For example, for the pairing of the first day and the second day, the intensity mean value difference is 142.7-145.2=-2.5, and the texture entropy value difference is 5.0-4.8=0.2. This process generates a two-dimensional difference vector for each time interval, and arranges all time interval difference vectors in order to obtain the adjacent time point feature difference sequence of the region unit.

[0139] According to the adjacent time point feature difference sequence of each region unit generated in the previous step, the system continues to process the data of each region unit between each continuous time point. Taking the data between the first day and the second day of region unit index 50 as an example, the feature difference vector is (-2.5, 0.2). The system first calculates the absolute values of the two difference components respectively to obtain (2.5, 0.2). Then, the sum of the two absolute values is 2.5+0.2=2.7. Then, the sum value is divided by the number of feature parameters that constitute the difference vector, that is, 2 (intensity mean value and texture entropy value), to calculate the average difference value of the feature parameters between the continuous time points, that is, 2.7 / 2=1.35. This value is the statistical difference of the region unit in this time interval. The system will repeat this calculation for all continuous time points of the region unit (the second day and the third day, the third day and the fourth day, etc.), generate a statistical difference value for each time interval, and finally generate a time series composed of multiple statistical difference values for the region unit, that is, the statistical difference of the continuous time points of the region unit.

[0140] Based on the statistical difference of each potential abnormal region unit calculated at the continuous time points of the region unit, the system begins to build the final evolution sequence data structure, which takes the region unit index as the unique key. The system will traverse all the indexes of the potential abnormal regions, such as indexes 50, 117, 189. For each index, the system will collect the statistical difference values calculated at all continuous time intervals, and arrange these values in a list in chronological order, for example, for index 50, the statistical difference sequence at continuous time points may be [1.35, 1.80, 0.95, 1.10, …, 2.50]. The system will take this value list as the value associated with the index key 50. After traversing all potential abnormal region units, a mapping set is finally formed, where each key is a region unit index, and the corresponding value is the complete time series data sequence of the feature difference metric of the region over time, which is the feature difference evolution sequence.

[0141] The acquisition step of the region evolution rate parameter is:

[0142] Based on the feature difference evolution sequence, the statistical difference of each region unit at continuous time points is extracted one by one based on the region unit index. According to the time label, the numerical difference between the feature difference statistical value of each region unit and the adjacent time label is calculated, and divided by the time interval to determine the initial rate of change of the difference of each region unit, and generate a set of feature difference initial rate values.

[0143] According to the set of feature difference initial rate values, the positive and negative signs of the initial rate values of the region unit feature difference are judged respectively. If the rate value is positive, it is marked that the feature difference of the region unit shows an increasing trend, and if the rate value is negative, it is marked that the feature difference of the region unit shows a decreasing trend, and a set of region unit difference evolution trend is obtained.

[0144] Based on the set of region unit difference evolution trends, the number of times of the increasing trend and the decreasing trend of the feature difference of each region unit in all continuous time periods is counted respectively, and the number of times is multiplied by the average absolute value of the rate of the corresponding region unit in the set of feature difference initial rate values to form the region evolution rate parameter.

[0145] Specifically, based on the feature difference evolution sequence, the system takes the regional unit index as the processing benchmark, and extracts the statistical difference time sequence of each potential abnormal regional unit one by one. Taking the regional unit index 50 as an example, its statistical difference sequence in the continuous 9 time intervals (for example, collected once a day, a total of 10 days of data) is [1.35, 1.80, 0.95, 1.10, 1.15, 0.80, 1.90, 2.10, 2.50]. The system will perform difference calculation on this sequence to obtain the change rate of the feature difference. Specifically, the system will calculate the difference between adjacent statistical difference values. For example, the first rate value is obtained by subtracting the first statistical difference value from the second statistical difference value, that is, 1.80-1.35=0.45. The second rate value is obtained by subtracting the second value from the third value, that is, 0.95-1.80=-0.85. In this way, the difference calculation of all adjacent values is completed. Considering that the collection time interval is fixed (for example, once a day, the time interval is 1 day), therefore, the difference value is the initial rate of change of each regional unit difference. The rate sequence [0.45, -0.85, 0.15, 0.05, -0.35, 1.10, 0.20, 0.40] calculated for the region 50 is associated with the region index. After traversing all potential abnormal regional units and repeating this process, the initial rate value set of the feature difference is formed.

[0146] According to the feature difference initial rate value set, the system will process the initial rate value sequence of each regional unit one by one to determine its evolution trend in each time interval. Taking the regional unit index 50 as an example, its rate sequence is [0.45, -0.85, 0.15, 0.05, -0.35, 1.10, 0.20, 0.40]. The system will traverse each rate value in this sequence and judge its positive or negative sign. For the first rate value 0.45, it is a positive number, which is marked as "increasing trend". For the second rate value -0.85, it is a negative number, which is marked as "decreasing trend". If the rate value is 0, it is marked as "no change trend". The system performs this judgment on all 8 rate values in the sequence, thereby generating a corresponding trend label sequence for region 50: ["increasing trend", "decreasing trend", "increasing trend", "increasing trend", "decreasing trend", "increasing trend", "increasing trend", "increasing trend"]. This process will be applied to the rate sequence of all potential abnormal regional units. The final generated set, with the regional index as the key and the trend label sequence as the value, is the regional unit difference evolution trend set.

[0147] Based on the set of regional unit difference evolution trend and the set of initial feature difference rate value, the system calculates a comprehensive evolution rate parameter for each regional unit, which aims to quantify the overall change intensity and direction of its feature difference. The specific calculation process is as follows. Taking regional unit index 50 as an example, the system first obtains its trend label sequence from the set of regional unit difference evolution trend, and counts the number of "increasing trend" and "decreasing trend" in it, obtaining 6 times of increasing trend and 2 times of decreasing trend. At the same time, the system obtains its rate sequence [0.45, -0.85, 0.15, 0.05, -0.35, 1.10, 0.20, 0.40] from the set of initial feature difference rate value, and calculates the average value of the absolute values of these rate values, i.e. (|0.45|+|-0.85|+|0.15|+|0.05|+|-0.35|+|1.10|+|0.20|+|0.40|) / 8=3.55 / 8≈0.444. Finally, the difference between the number of increasing trend and the number of decreasing trend is multiplied by the average absolute value of the rate to form the regional evolution rate parameter, and the calculation formula is: P m =(C inc,m -C dec,m )×Avg(|r m |), where P m is the regional evolution rate parameter of regional unit m, C inc,m and C dec,m are the number of increasing and decreasing trends of the region respectively, and Avg(|r m |) is the average value of the absolute values of all initial rates of the region. For regional unit 50, the regional evolution rate parameter is (6-2)×0.444=4×0.444=1.776.

[0148] The acquisition steps of photovoltaic component defect early warning judgment are as follows:

[0149] Based on the regional evolution rate parameter, the evolution rate and the corresponding number of time intervals are extracted one by one according to the regional unit index. According to the statistical value of the feature difference at the latest time as the current starting value, the evolution rate value of each regional unit is multiplied by the time interval number of the prediction step, and is added to the current starting value to generate a set of feature difference measurement prediction values of each regional unit in the future period.

[0150] According to the set of feature difference measurement prediction values of each regional unit in the future period, the prediction value of each regional unit is compared with the maximum value, the average value and the upper threshold of the stable interval of the feature difference measurement in the historical period in turn. If the prediction value exceeds the maximum value, the average value or the upper threshold of the stable interval, it is marked as having a future feature abnormal trend, otherwise it is marked as being within the acceptable range of feature change, to obtain the regional unit feature difference out-of-limit judgment label.

[0151] Based on the regional unit feature difference over-limit judgment label, all regional units with future feature abnormal trend are indexed and merged, and a defect identification table is generated by combining the photovoltaic module number and the regional unit index to generate a photovoltaic module defect early warning judgment.

[0152] Specifically, based on the regional evolution rate parameter, the system processes each potential abnormal regional unit one by one. Taking the regional unit index 50 as an example, the system first extracts its regional evolution rate parameter, which is 1.776, and obtains the latest feature difference statistical value from the feature difference evolution sequence [1.35, 1.80, 0.95, 1.10, 1.15, 0.80, 1.90, 2.10, 2.50], which is 2.50, as the predicted current starting value. Then, according to the statistical analysis of the development speed of the same type of photovoltaic module defects and the maintenance response cycle of the operation and maintenance department, a fixed prediction step is set, for example, according to experience, the defect from the initial appearance to the maintenance usually has a window period of about 7 days, in order to leave response time, the prediction step is set to 3 time intervals (representing 3 days in the future), next, the system multiplies the regional evolution rate parameter 1.776 of the regional unit by the prediction step 3, and obtains the evolution increment 5.328, finally, the increment is added to the current starting value 2.50, and the feature difference measure prediction value of the regional unit in the future period is calculated as 2.50+5.328=7.828, this process will be applied to all potential abnormal regions, and finally a set of feature difference measure prediction values of each regional unit in the future period is generated.

[0153] According to the set of predicted values of the feature difference metric of each regional unit in the future time period, the system performs multi-dimensional risk assessment on the predicted value of each regional unit. Taking regional unit index 50 as an example, the predicted value is 7.828. The system first calculates three key decision criteria from the historical feature difference evolution sequence [1.35, 1.80, 0.95, 1.10, 1.15, 0.80, 1.90, 2.10, 2.50] of the regional unit: the historical maximum value is 2.50, the historical average value is 1.51, and the upper threshold of the stable interval is calculated as follows: first, calculate the standard deviation of the sequence, which is about 0.562, then add 1.5 times (i.e. 0.843) of the standard deviation to the historical average value to obtain the upper threshold of the stable interval, which is 1.51 + 0.843 = 2.353. This threshold represents the statistical normal upper limit of the historical fluctuations of the region. Subsequently, the system compares the predicted value 7.828 with the three benchmark values to determine whether any of the following conditions is met: the predicted value is greater than the historical maximum value (7.828 > 2.50), or the predicted value is greater than the historical average value (7.828 > 1.51), or the predicted value is greater than the upper threshold of the stable interval (7.828 > 2.353). Since all conditions are met in this case, the system labels regional unit 50 as “future feature abnormal trend exists”. If the predicted value of another region does not exceed any of its corresponding benchmark values, it is labeled as “within acceptable range of feature change”. After completing the determination of all regions, the regional unit feature difference out-of-limit determination label is obtained.

[0154] Based on the regional unit feature difference out-of-limit determination label, the system performs the final warning information collection and generation. First, the system filters out the indexes of all regional units labeled as “future feature abnormal trend exists”, for example, after the last round of determination, regional units with indexes 50 and 189 are identified. Then, the system retrieves the unique identifier of the photovoltaic module being analyzed, i.e. the photovoltaic module number, for example “PV-Panel-SN12345”. Then, the system creates a detailed defect record for each abnormal regional unit selected, which is organized into a structured entry containing the photovoltaic module number, regional unit index, regional evolution rate parameter of the region, predicted value of the feature difference metric in the future time period, etc. For example, the record generated for regional unit 50 is {photovoltaic module number: “PV-Panel-SN12345”, regional unit index: 50, regional evolution rate parameter: 1.776, predicted value of feature difference metric: 7.828}. All such records are collected to form the final defect identification table, which is the photovoltaic module defect warning determination generated by this detection.

[0155] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application in other forms. Any person skilled in the art can make changes or modifications to the above disclosed technical contents 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 according to the technical essence of the present application, without departing from the technical solution content of the present application, still belongs to the protection scope of the technical solution of the present application.

Claims

1. A photovoltaic module defect detection method based on image recognition, characterized in that: The following steps are involved: Obtain a photovoltaic module image of the photovoltaic module to be inspected, calculate the intensity mean and texture entropy value for predefined regional units in the photovoltaic module image, combine the intensity mean and texture entropy value of each region into a feature vector, and establish a regional dual feature vector set; Based on the regional dual feature vector set, generating a feature mapping point cloud according to the spatial neighborhood features, judging the point cloud distribution density based on the feature mapping point cloud, judging abnormal points, and obtaining a potential abnormal feature point set; Photovoltaic module images of the same photovoltaic module at different times are obtained. For the area corresponding to the potential abnormal feature point set, the intensity mean and texture entropy value are extracted from the photovoltaic module image at each time point to construct a multi-time point feature value column. Based on the multi-time point feature value column, the statistical difference of the feature value between consecutive time points is calculated to obtain a feature difference evolution sequence; Based on the characteristic difference evolution sequence, the rate and direction of change of the characteristic difference measurement of each region over time are analyzed, and regional evolution rate parameters are established. Based on the regional evolution rate parameters, the characteristic difference measurement of future time periods is calculated to generate photovoltaic module defect warning judgments.

2. The photovoltaic module defect detection method based on image recognition according to claim 1, characterized in that: The steps for obtaining the regional dual feature vector set are: Acquire a photovoltaic module image of the photovoltaic module to be inspected, mark a number of predefined area units in the photovoltaic module image, and extract a pixel set of each predefined area unit to obtain a regional pixel set; Based on the regional pixel sets, averaging the grayscale values ​​of all pixels in each predefined regional unit to generate an intensity mean of the corresponding regional unit, and simultaneously calculating the entropy value of the grayscale value distribution of each regional pixel set to form a texture entropy value of the corresponding regional unit; Based on the intensity mean value of the corresponding regional unit and the texture entropy value of the corresponding regional unit, a regional dual feature vector set is formed.

3. The photovoltaic module defect detection method based on image recognition according to claim 1, characterized in that: The steps for obtaining the feature mapping point cloud are: Based on the regional dual feature vector set, the intensity mean and texture entropy value of each regional unit in the current image are extracted, and the units are numbered in regional order to identify the spatial coordinates of each regional unit. Subsequently, the image data of the photovoltaic module at the same spatial position in the previous three days are retrieved, and the intensity mean and texture entropy value of the regional unit with the same number in the images of the previous three days are extracted to generate a three-time point historical feature sequence group; Calculating a structural disturbance index based on the three-time point historical feature sequence group and combining the spatial neighborhood features of each regional unit in the current image; Based on the structural perturbation index, the structural perturbation indexes of all area units in the current image are sorted in ascending order according to their size, and the area units whose structural perturbation indexes are in the top 25% of all area units are selected. The feature vectors of the area units are combined into a target mapping vector set to generate a feature mapping point cloud.

4. The photovoltaic module defect detection method based on image recognition according to claim 1, characterized in that: The steps for obtaining the potential abnormal feature point set are: Based on the feature map point cloud, the two-dimensional coordinate value of each area unit in the point cloud is read, a fixed radius is set as the search range, all point cloud coordinate pairs are traversed, the number of point pairs forming a geometric neighborhood relationship within the fixed radius is identified, the number of neighboring points of each point is obtained, and a neighborhood coordinate set is constructed; Calculate the density outlier index of each point based on the neighborhood coordinate set and the number of neighboring points; Based on the density outlier index, all regional units are sorted in descending order according to the size of the density outlier index, and regional units whose density outlier index is higher than the median of the outlier index of all regional units are screened to generate a set of potential abnormal feature points.

5. The photovoltaic module defect detection method based on image recognition according to claim 1, characterized in that: The steps for obtaining the multi-time point feature value column are: Acquire PV module images taken at different acquisition times of the same PV module, parse the image time tags, and extract the pixel data of the corresponding regional unit from each PV module image based on the regional unit index information recorded in the potential abnormal feature point set to form a multi-time pixel set of the corresponding regional unit; Based on the corresponding regional unit multi-time pixel set, calculating the average value of the pixel grayscale value and the texture entropy value of the grayscale distribution for each pixel and each regional unit, recording and pairing them into regional unit feature parameter combinations in sequence, arranging the regional unit feature parameter combinations in order of time tags, and generating a regional unit multi-time feature parameter combination sequence; Based on the multi-time characteristic parameter combination sequence of the regional unit, the multi-time characteristic parameter combination sequences of all regional units are aggregated to generate a multi-time point characteristic value column.

6. The photovoltaic module defect detection method based on image recognition according to claim 1, characterized in that: The steps for obtaining the characteristic differential evolution sequence are: Based on the multi-time point feature value sequence, the intensity mean and texture entropy value of each regional unit at each time point are extracted according to the regional unit index, the feature parameters of adjacent time points are paired one by one, and the difference between the intensity mean and texture entropy value between adjacent time points is calculated to obtain the adjacent time point feature difference sequence of each regional unit; According to the sequence of feature difference values ​​at adjacent time points, the absolute values ​​of the intensity mean and texture entropy value differences of each regional unit between consecutive time points are calculated respectively, the absolute values ​​of the intensity mean and texture entropy value differences of the same regional unit are summed, and then divided by the number of feature parameters at corresponding consecutive time points, and the average difference value of the feature parameters at consecutive time points is calculated to obtain the statistical difference of the consecutive time points of the regional unit; Based on the statistical differences of the regional units at consecutive time points, the statistical difference values ​​are recorded one by one in the order of the regional unit indexes, and with the regional units as the index keys, a time series data sequence is established to record the changes in the statistical differences of the characteristic parameters of all regional units at consecutive time points, thereby generating a characteristic difference evolution sequence.

7. The photovoltaic module defect detection method based on image recognition according to claim 1, characterized in that: The steps for obtaining the regional evolution rate parameter are: Based on the characteristic difference evolution sequence, the statistical differences of each regional unit at consecutive time points are extracted one by one with the regional unit index as the benchmark. According to the time tag, the numerical difference of the characteristic difference statistical value of each regional unit between adjacent time tags is calculated, and the difference is divided by the time interval to determine the initial rate of difference change of each regional unit, and generate a set of characteristic difference initial rate values; According to the characteristic difference initial rate value set, the positive and negative signs of the regional unit characteristic difference initial rate values ​​are determined respectively. If the rate value is positive, it indicates that the regional unit characteristic difference is increasing. If the rate value is negative, it indicates that the regional unit characteristic difference is decreasing. Thus, a regional unit difference evolution trend set is obtained. Based on the set of regional unit difference evolution trends, the number of occurrences of the increasing trend and decreasing trend of characteristic differences in each regional unit in all continuous time periods is counted respectively, and the number of occurrences is multiplied by the average absolute value of the corresponding regional unit rate in the set of characteristic difference initial rate values ​​to form a regional evolution rate parameter.

8. The photovoltaic module defect detection method based on image recognition according to claim 1, characterized in that: The steps for obtaining the photovoltaic module defect early warning judgment are as follows: Based on the regional evolution rate parameter, the evolution rate and the corresponding time interval number are extracted one by one according to the regional unit index. The characteristic difference statistical value of the most recent time is used as the current starting value. The evolution rate value of each regional unit is multiplied by the number of time intervals of the prediction step, and the result is superimposed on the current starting value to generate a set of characteristic difference metric prediction values ​​for each regional unit in the future period. Based on the predicted value set of the characteristic difference metric of each regional unit in the future time period, the predicted value of each regional unit is compared with the maximum value, average value and upper limit threshold of the characteristic difference metric in the historical time period in turn. If the predicted value exceeds the maximum value, average value or upper limit threshold of the stable interval, it is marked as having a future characteristic abnormal trend; otherwise, it is marked as being within the acceptable range of characteristic change, and a regional unit characteristic difference exceeding limit judgment label is obtained; Based on the regional unit characteristic difference exceeding limit judgment label, all regional unit indexes with future characteristic abnormality trends are unified and merged, and a defect identification table is generated by combining the photovoltaic module number and the regional unit index to generate a photovoltaic module defect warning judgment.

9. The photovoltaic module defect detection system according to any one of claims 1 to 8, wherein: include: An image feature extraction module obtains a photovoltaic module image of the photovoltaic module to be inspected, calculates the intensity mean and texture entropy value for predefined regional units in the photovoltaic module image, and combines the intensity mean and texture entropy value of each region into a feature vector to establish a regional dual feature vector set; A feature point cloud generation and anomaly detection module generates a feature mapping point cloud based on the regional dual feature vector set and the spatial neighborhood features, determines the point cloud distribution density and abnormal points based on the feature mapping point cloud, and obtains a potential abnormal feature point set; A time series feature difference analysis module obtains photovoltaic module images of the same photovoltaic module at different times, extracts the intensity mean and texture entropy value from the photovoltaic module image at each time point in the area corresponding to the potential abnormal feature point set, constructs a multi-time point feature value column, and calculates the statistical difference of the feature values ​​between consecutive time points based on the multi-time point feature value column to obtain a feature difference evolution sequence; The defect evolution trend prediction module analyzes the rate and direction of change of the characteristic difference measurement of each region over time based on the characteristic difference evolution sequence, establishes regional evolution rate parameters, and calculates the characteristic difference measurement of future time periods based on the regional evolution rate parameters to generate photovoltaic module defect warning judgments.

Images