A hot spot detection method and system based on adaptive B-spline function curve fitting

Abstract

The present invention belongs to the field of image processing technology and discloses a hot spot detection method and system using adaptive B-spline function curve fitting. The method uses a Gaussian filter to filter and preprocess the collected infrared image of a photovoltaic module, identifies and determines the grayscale histogram of its image pixels, establishes an image grayscale histogram, and uses an adaptive B-spline function to fit the grayscale histogram of the photovoltaic module infrared thermal image to segment the hot spots, extracts features from the infrared thermal image, constructs data samples, trains and tests a decision tree classifier based on the data samples, and finally identifies the hot spots formed on the photovoltaic module at the system application end through the decision tree classifier. This method can quickly identify the cause of the hot spots and improve diagnostic efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a hot spot detection method and system using adaptive B-spline function curve fitting. Background Art

[0002] Single solar cells cannot be used directly as power sources; to use them as power sources, several single cells must be connected in series and parallel and tightly sealed into modules. Solar cell modules are the core and most important part of solar power generation systems. Their function is to convert solar energy into electrical energy, or send it to batteries for storage, or drive loads.

[0003] The hot spots formed on the system application end of photovoltaic modules come from two aspects: shading and battery defects. Shading comes from bird droppings, fallen leaves, snow residue, dust, clouds, as well as plants, buildings, adjacent string modules, etc. When shading occurs and the bypass diode is not turned on, the shaded cells in the module or the shaded cells are partially in a "reverse bias" state, generating heat under the action of the system current, forming hot spots; battery defects include: excessive leakage current or concentrated leakage area, excessive string resistance, small parallel resistance, hidden cracks, cracks, edge short circuits, power mixing, black chips, sintering short circuits, cold solder joints, etc., which may cause modules to produce hot spots on the system application end.

[0004] In the existing technology, it is impossible to perform abnormal diagnosis based on the hot spots formed on the system application side, and it can only be identified manually, resulting in slow response and cumbersome processing. Summary of the Invention

[0005] The purpose of the present invention is to provide a hot spot detection method and system based on adaptive B-spline function curve fitting to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A hot spot detection method based on adaptive B-spline function curve fitting, comprising the steps of:

[0008] Collect infrared thermal images of photovoltaic modules under different environmental conditions;

[0009] Preprocessing the infrared thermal image to obtain a grayscale image, and performing statistics on the grayscale image to establish a grayscale histogram;

[0010] Establishing n grayscale intervals that evenly divide the grayscale value range, and calculating the grayscale sum of each grayscale interval of the infrared thermal image;

[0011] Determine the grayscale interval with the maximum grayscale sum, and mark the grayscale value corresponding to the center point of the grayscale interval as max;

[0012] The grayscale value range is divided based on the dividing point max, and p and q grayscale sub-intervals are established in the grayscale interval on both sides after the division to construct the node vector;

[0013] Construct a B-spline function least squares fitting curve based on the node vectors, determine the peak with the highest grayscale value of the least squares fitting curve, and set the grayscale value of the first trough after the peak as the image segmentation threshold to obtain a hot spot segmentation map, and mark the hot spot image to obtain a label matrix N;

[0014] Extracting geometric characteristic parameters of each hot spot target area through the hot spot segmentation map, wherein the geometric characteristic parameters include area characteristics, perimeter characteristics and shape factor characteristics;

[0015] Construct data samples based on geometric feature parameters, divide them into training sets and test sets, and train and test the decision classifier;

[0016] Based on the trained decision classifier, hot spots are identified and the fault type is determined.

[0017] As a further solution of the present invention: in the step of pre-processing the infrared thermal image to obtain a grayscale image, a Gaussian filter is selected to perform filtering pre-processing on the infrared image of the photovoltaic module;

[0018] In the step of performing statistics on the grayscale image and establishing a grayscale histogram, the grayscale values ​​of the image pixels are marked based on the preprocessed grayscale image, and the pixel statistics of the grayscale image are performed based on the grayscale values ​​to establish a grayscale histogram. The statistical equation is:

[0019] ;

[0020] in, Represents the grayscale value in the grayscale image, Indicates the total number of pixels corresponding to each grayscale value.

[0021] As a further solution of the present invention: in the step of marking the grayscale values ​​of image pixels based on the preprocessed grayscale image, a scanning marking method is adopted in which the grayscale image is scanned column by column and pixel by pixel from left to right and from top to bottom to determine its grayscale value and mark the grayscale value.

[0022] As a further solution of the present invention: in the step of segmenting the grayscale value range based on the dividing point max, and establishing p and q grayscale sub-intervals in the grayscale interval on both sides after segmentation, respectively, to construct a node vector;

[0023] The total grayscale value interval is [0, 256]. The grayscale interval [0, max] on the left side of the dividing point is divided into p grayscale subintervals, and the grayscale interval [max, 256] on the right side of the dividing point is divided into q grayscale subintervals. The node vector is:

[0024] [0,0,0,max / p,2max / p,…,(p-1)max / p,max,max+(255-max) / q,…,max+(q-1)(255-max) / q,255,255,255].

[0025] As a further solution of the present invention: in the step of constructing the B-spline function least square fitting curve according to the node vector, the 256 sample points of the grayscale histogram are fitted by the adaptive B-spline function least square method. Fitting is performed, and the curve function after fitting is:

[0026] ;

[0027] The selection of the threshold depends on the curve fitting accuracy of the [max, 256] grayscale interval, so the q value is set larger than the p value.

[0028] As a further solution of the present invention: in the step of extracting the geometric feature parameters of each hot spot target area through the hot spot segmentation map, the step of calculating the area feature of the hot spot area is: traversing each hot spot target area image, counting the total number of pixels contained in each hot spot target area image, and obtaining the area feature.

[0029] As a further solution of the present invention: in the step of extracting the geometric feature parameters of each hot spot target area through the hot spot segmentation map, the step of calculating the perimeter feature of the hot spot area is: traversing the pixels of the hot spot area image in the hot spot segmentation map, counting the number of all pixels located at the edge of the hot spot area, and obtaining the perimeter feature of the hot spot.

[0030] As a further solution of the present invention: in the step of extracting the geometric characteristic parameters of each hot spot target area through the hot spot segmentation map, in the step of calculating the shape factor characteristics of the hot spot target area image, the shape factor characteristics are obtained by the formula:

[0031] ;

[0032] in, is the shape factor feature of the nth hot spot target area image, is the area feature of the nth hot spot target area image, and G(n) is the perimeter feature of the nth hot spot target area image.

[0033] As a further solution of the present invention: the step of training the decision classifier specifically includes: calculating the optimal division point of each continuous attribute, determining the optimal sample set according to the optimal division point, calculating the information gain corresponding to the optimal sample set, calculating the information gain rate corresponding to each continuous attribute, selecting the optimal division attribute according to the information gain rate, and generating a decision classifier.

[0034] The embodiment of the present invention aims to provide a hot spot detection system using adaptive B-spline function curve fitting, comprising:

[0035] Image acquisition module, used to collect infrared thermal images of photovoltaic modules under different environmental conditions;

[0036] A grayscale processing module pre-processes the infrared thermal image to obtain a grayscale image, and performs statistics on the grayscale image to establish a grayscale histogram;

[0037] An interval calculation module is configured to establish n grayscale intervals that evenly divide the grayscale value range, and calculate the grayscale sum of each grayscale interval of the infrared thermal image;

[0038] A marking selection module determines the grayscale interval with the largest grayscale sum, and marks the grayscale value corresponding to the center point of the grayscale interval as max;

[0039] The node vector construction module divides the grayscale value range based on the dividing point max, and establishes p and q grayscale sub-intervals in the grayscale interval on both sides of the segmentation to construct the node vector;

[0040] The function fitting segmentation module constructs a B-spline function least squares fitting curve based on the node vector, determines the peak with the highest grayscale value of the least squares fitting curve, and sets the grayscale value of the first trough after the peak as the image segmentation threshold to obtain a hot spot segmentation map;

[0041] A feature extraction module extracts geometric feature parameters of each hot spot target area through the hot spot segmentation map, wherein the geometric feature parameters include area feature, perimeter feature and shape factor feature;

[0042] The classification model construction module constructs data samples based on geometric feature parameters, divides the data into training sets and test sets, and trains and tests the decision classifier;

[0043] The abnormality judgment module identifies hot spots based on the trained decision classifier and determines the fault type.

[0044] Compared with the prior art, the beneficial effects of the present invention are: by selecting a Gaussian filter to filter and preprocess the collected infrared image of the photovoltaic module, the grayscale histogram of its image pixel is identified and determined, the image grayscale histogram is established, and the adaptive B-spline function is used to fit the grayscale histogram of the infrared thermal image of the photovoltaic module to segment the hot spots, so as to extract features of the infrared thermal image, construct a data sample, train and test a decision classifier based on the data sample, and finally identify the hot spots formed on the photovoltaic module at the system application end through the decision classifier, which can quickly identify the cause of the hot spots and improve the diagnostic efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The figure is a flow chart of a hot spot detection method based on adaptive B-spline function curve fitting.

[0046] Figure 2 Schematic diagram of grayscale image in a hot spot detection method based on adaptive B-spline function curve fitting.

[0047] Figure 3 Grayscale histogram of a hot spot detection method based on adaptive B-spline function curve fitting.

[0048] Figure 4 Schematic diagram of the fitting curve in a hot spot detection method using adaptive B-spline function curve fitting.

[0049] Figure 5 This is a hot spot segmentation map in a hot spot detection method based on adaptive B-spline function curve fitting.

[0050] Figure 6 A block diagram of a hot spot detection system based on adaptive B-spline function curve fitting. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to 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.

[0052] The specific implementation of the present invention is described in detail below with reference to specific embodiments.

[0053] like Figure 1 The hot spot detection method using adaptive B-spline function curve fitting provided in one embodiment of the present invention includes the following steps:

[0054] S10, collecting infrared thermal images of photovoltaic modules under different environmental conditions;

[0055] In this step, infrared thermal images of the photovoltaic components under different environmental conditions are collected, that is, infrared thermal imaging equipment is used to collect images of the photovoltaic components to obtain infrared thermal images.

[0056] S20, preprocessing the infrared thermal image to obtain a grayscale image, and performing statistics on the grayscale image to establish a grayscale histogram;

[0057] In this step, the infrared image of the photovoltaic module is filtered and preprocessed by Gaussian filtering and grayscale conversion, so that the following is obtained: Figure 2 The grayscale image shown in , and the total number of pixels with different grayscale values ​​in the image are counted to obtain Figure 3 The histogram of grayscale values ​​is shown.

[0058] S30, establishing n grayscale intervals that evenly divide the grayscale value range, and calculating the grayscale sum of each grayscale interval of the infrared thermal image;

[0059] In this embodiment, the grayscale value (0-255) is evenly divided into n grayscale intervals, and the grayscale sum of each grayscale interval of the infrared thermal image of the photovoltaic module is calculated. Here, n is generally set to be greater than or equal to 10. Here, n is set to be 10. Therefore, the program code used to calculate the grayscale sum is as follows:

[0060] 1 [counts,x]=imhist(img);

[0061] 2 sum=zeros(1,10);

[0062] 3 sumx=zeros(1,10);

[0063] 4 for j=1:10

[0064] 5 for i=((j-1)*25+1):(25*j)

[0065] 6 sum(j)=sum(j)+counts(i);

[0066] 7 end

[0067] 8 sumx(j)=25*j-12;

[0068] 9 end.

[0069] S40, determining the grayscale interval with the maximum grayscale sum, and marking the grayscale value corresponding to the center point of the grayscale interval as max;

[0070] In this step, the grayscale sum of different grayscale intervals is judged, and the grayscale interval with the largest grayscale value is selected and its midpoint is marked as max (all integers, if the interval span is 1, it represents the value). The program code for finding max is as follows:

[0071] 1 maxsum=0;

[0072] 2 maxx=0;

[0073] 3 for j=1:10

[0074] 4 if sum(j)>=maxsum

[0075] 5 maxsum=sum(j);

[0076] 6 maxx=sumx(j);

[0077] 7 end

[0078] 8 end.

[0079] S50, dividing the grayscale value range based on the dividing point max, and establishing p and q grayscale sub-intervals in the grayscale interval on both sides after the division, respectively, to construct a node vector;

[0080] In this embodiment, since the selection of the threshold in the hot spot segmentation method based on the grayscale histogram adaptive node B spline function curve fitting mainly depends on the curve fitting accuracy of the [max, 255] grayscale interval, p is generally set to a smaller value and q is set to a larger value during setting. Here, p=3 and q=7 are set.

[0081] S60, constructing a B-spline function least squares fitting curve based on the node vector, determining the peak with the highest grayscale value of the least squares fitting curve, and setting the grayscale value of the first trough after the peak as the image segmentation threshold to obtain a hot spot segmentation map, and marking the hot spot image to obtain a labeling matrix N.

[0082] In this embodiment, Figure 4 The curve obtained by fitting is shown below. The peak with the highest gray value of the adaptive B-spline fitting curve of the gray histogram is found, and the gray value corresponding to the first trough after the peak is determined as the image segmentation threshold. Finally, the hot spot segmentation map is obtained as shown below. Figure 5 As shown, the program code for determining the threshold is:

[0083] 1 max=0;

[0084] 2 top=0;

[0085] 3 for i=0:255

[0086] 4 if vals(i+1)>max

[0087] 5 max=vals(i+1);

[0088] 6 top=i+1;

[0089] 7 end

[0090] 8 end

[0091] 9 for i=top:254

[0092] 10 if vals(i+1)>vals(i)

[0093] 11 level2=i;

[0094] 12 break;

[0095] 13 end

[0096] 14 end;

[0097] Furthermore, in the process of marking the hot spot image, the pixel points of the hot spot image are , and its corresponding label , The initial value is 1. Scan the hot spot segmentation map to be processed from left to right and from top to bottom. During this process, the number of each hot spot target area image and the pixel points contained in the hot spot image corresponding to each number are determined according to this rule;

[0098] (1) If , , then let

[0099] (2) If , , then let

[0100] (3) If , ,and , then let

[0101] (4) If , ,but

[0102] (5) If , ,but ,if ,but , ;if ,but , .

[0103] S70, extracting geometric feature parameters of each hot spot target area through the hot spot segmentation map, wherein the geometric feature parameters include area feature, perimeter feature and shape factor feature.

[0104] In this step, the geometric characteristic parameters of the hot spot target area are calculated one by one. Specifically, the area characteristic, perimeter characteristic and shape factor characteristic are calculated:

[0105] (1) Area characteristics A of the hot spot target area:

[0106] In the previous step, the label matrix N of the hot spot segmentation map P has been obtained. Traverse the matrix N. If ,but , and finally obtained That is the area of ​​hot spot No. 1, The initial value is 0;

[0107] Traverse the matrix N, if ,but , and finally obtained That is the area of ​​the hot spot area n, The initial value is 0;

[0108] (2) Perimeter feature G of the hot spot target area:

[0109] if ,and If at least one of the four fields has a value of 0, then , and finally obtained is the perimeter of the nth hot spot area, The initial value is 0;

[0110] (3) Shape factor characteristics C of the hot spot target area:

[0111] For the hot spot target area n, calculate , It is the shape factor of the target area of ​​the hot spot No. n, which can describe the degree to which the target shape is close to a circle.

[0112] S80: constructing data samples according to the geometric feature parameters, dividing the data into a training set and a test set, and training and testing the decision classifier.

[0113] In this step, for constructing data samples:

[0114] Hot spots formed by photovoltaic modules at the system application end come from two aspects: obstruction and cell defects. Obstruction comes from bird droppings, fallen leaves, snow residue, dust, clouds, as well as plants, buildings, adjacent string modules, etc. Cell defects include: excessive leakage current or concentrated leakage area, excessive string resistance, insufficient parallel resistance, hidden cracks, splinters, edge short circuits, power mixing, black chips, sintered short circuits, cold solder joints, etc. Therefore, according to the cause of hot spots, hot spots are divided into two categories: external obstruction faults and internal defect faults. In the data table, we define external obstruction faults as 1 and internal defect faults as 2, and establish a hot spot fault classification data sample. The example is as follows:

[0115] Hot spot type Hot spot area Hot spot circumference Hot spot shape factor 1 187 55.947 0.94185 2 30186 1291.288 0.977846 1 121 39.713 0.862373 1 120 39.068 0.861561 1 555 148.828 0.993591 1 112 37.108 0.84076 2 36774 1302.836 0.976155 2 22671 1274.324 0.983734 2 22676 1288.486 0.984425 1 250 56.038 0.037891 1 128 37.522 0.460749 1 6511 549.724 0.987718 1 182 54.839 0.9405 1 244 53.64 0 1 135 38.928 0.493505 1 6406 556.614 0.988119 2 31365 1255.256 0.976397 2 27146 1224.364 0.976459 1 1360 183.656 0.94719 1 1109 143.322 0.88737 1 69 26.2 0 1 1446 169.356 0.937213 1 5276 533.588 0.986485 1 120 36.298 0.523248 1 120 36.298 0.523248 1 127 37.522 0.439413 1 194 50.804 0.810375 1 561 89.616 0.653929 1 589 107.884 0.944333 1 817 109.538 0.81576 1 387 77.094 0.838922 1 244 53.64 0 1 884 113.226 0.798052 2 38876 1161.372 0.96957 2 20916 829.906 0.964677 1 120 36.298 0.523248 2 17598 921.216 0.959134 2 1591 180.934 0.868726 2 19942 1123.196 0.973724 2 5420 311.278 0.855435 2 44212 1375.792 0.981562 1 774 140.816 0.944028 2 9472 451.65 0.846087 2 5858 312.164 0.632586 2 4038 259.616 0.738662 2 23845 1023.13 0.947012 2 25558 1366.15 0.977268

[0116] For dividing the training set and test set, train and test the decision classifier:

[0117] The data samples are randomly divided into training set and test set, and the decision classifier is trained using the training set samples. The specific steps are as follows:

[0118] (1) The area, perimeter, and shape factor of the hot spot in the infrared thermal image segmentation of photovoltaic modules are all continuous values, not discrete quantities. Therefore, the bi-partition method is used to process the continuous attributes.

[0119] In this embodiment, taking the area feature as an example, the hot spot area is defined as a continuous attribute , It appeared on different values, sort these values ​​from small to large, and record them as , dividing point You can Divide into subsets and ,in Contains those in the attribute The upper value is not greater than A sample of Contains those on the attribute Upper value greater than Sample, select the interval , The median point As a candidate partition point, for continuous attributes , need to examine including The set of candidate partition points ,calculate (The sample set is based on the partition point Information gain after binary division), so that Maximized is the dividing point to be selected; among them,

[0120]

[0121] ;

[0122] (2) Calculate the information gain rate and select the optimal partition attribute:

[0123] The information gain rate is calculated as follows:

[0124] Assume that the proportion of the k-th class samples in the current sample set D is P k , then the information entropy of D is defined as:

[0125]

[0126] The smaller the value of , the higher the purity of D;

[0127] Assuming discrete attributes There are V possible values If you use To divide the sample set, V branch nodes will be generated, among which The branch nodes contain all the attributes in D The upper value is The sample is recorded as , so we can calculate the discrete attributes The "information gain" obtained by dividing the sample set:

[0128] ,

[0129] Then calculate the information gain rate:

[0130]

[0131] in,

[0132]

[0133] Ultimately, the choice makes The largest a is taken as the optimal partition attribute; we get Figure 3 The photovoltaic fault decision classifier shown; the code is as follows:

[0134] Decision tree for classification

[0135] 1if x1<2742 then node 2 elseif x1>=2742 then node 3 else 1

[0136] 2class = 1

[0137] 3if x3<0.985455 then node 4 elseif x3>=0.985455 then node 5 else 2

[0138] 4class = 2

[0139] 5 class = 1.

[0140] S90: Identify the hot spot based on the trained decision classifier and determine the fault type.

[0141] As another preferred embodiment of the present invention, in the step of pre-processing the infrared thermal image to obtain a grayscale image, a Gaussian filter is selected to perform filtering pre-processing on the infrared image of the photovoltaic module;

[0142] In the step of performing statistics on the grayscale image and establishing a grayscale histogram, the grayscale values ​​of the image pixels are marked based on the preprocessed grayscale image, and the pixel statistics of the grayscale image are performed based on the grayscale values ​​to establish a grayscale histogram. The statistical equation is:

[0143] ;

[0144] in, Represents the grayscale value in the grayscale image, Indicates the total number of pixels corresponding to each grayscale value.

[0145] As another preferred embodiment of the present invention, in the step of marking the grayscale values ​​of image pixels based on the preprocessed grayscale image, a scanning marking method is adopted in which the grayscale image is scanned column by column and pixel by pixel from left to right and from top to bottom to determine its grayscale value and mark the grayscale value.

[0146] As another preferred embodiment of the present invention, in the step of segmenting the grayscale value range based on the dividing point max, and establishing p and q grayscale sub-intervals in the grayscale intervals on both sides after segmentation, respectively, to construct a node vector;

[0147] The total grayscale value interval is [0, 256]. The grayscale interval [0, max] on the left side of the dividing point is divided into p grayscale subintervals, and the grayscale interval [max, 256] on the right side of the dividing point is divided into q grayscale subintervals. The node vector is:

[0148] [0,0,0,max / p,2max / p,…,(p-1)max / p,max,max+(255-max) / q,…,max+(q-1)(255-max) / q,255,255,255].

[0149] As another preferred embodiment of the present invention, in the step of constructing a B-spline function least squares fitting curve according to the node vector, the 256 sample points of the grayscale histogram are fitted by the adaptive B-spline function least squares method. Fitting is performed, and the curve function after fitting is:

[0150] ;

[0151] The selection of the threshold depends on the curve fitting accuracy of the [max, 256] grayscale interval, so the q value is set larger than the p value.

[0152] As another preferred embodiment of the present invention, in the step of extracting the geometric feature parameters of each hot spot target area through the hot spot segmentation map, the step of calculating the area feature of the hot spot area is: traversing each hot spot target area image, counting the total number of pixels contained in each hot spot target area image, and obtaining the area feature.

[0153] As another preferred embodiment of the present invention, in the step of extracting the geometric feature parameters of each hot spot target area through the hot spot segmentation map, the step of calculating the perimeter feature of the hot spot area is: traversing the pixels of the hot spot area image in the hot spot segmentation map, counting the number of all pixels located at the edge of the hot spot area, and obtaining the perimeter feature of the hot spot.

[0154] As another preferred embodiment of the present invention, in the step of extracting the geometric characteristic parameters of each hot spot target area through the hot spot segmentation map, in the step of calculating the shape factor characteristics of the hot spot target area image, the shape factor characteristics are obtained by the formula:

[0155] ;

[0156] in, is the shape factor feature of the nth hot spot target area image, is the area feature of the nth hot spot target area image, and G(n) is the perimeter feature of the nth hot spot target area image.

[0157] As another preferred embodiment of the present invention, the step of training the decision classifier specifically includes: calculating the optimal division point of each continuous attribute, determining the optimal sample set according to the optimal division point, calculating the information gain corresponding to the optimal sample set, calculating the information gain rate corresponding to each continuous attribute, selecting the optimal division attribute according to the information gain rate, and generating a decision classifier.

[0158] like Figure 6 As shown, the present invention also provides a hot spot detection system with adaptive B-spline function curve fitting, which includes:

[0159] Image acquisition module 100, used to collect infrared thermal images of photovoltaic modules under different environmental conditions;

[0160] A grayscale processing module 200 pre-processes the infrared thermal image to obtain a grayscale image, and performs statistics on the grayscale image to establish a grayscale histogram;

[0161] An interval calculation module 300 establishes n grayscale intervals that evenly divide the grayscale value range, and calculates the grayscale sum of each grayscale interval of the infrared thermal image;

[0162] The marking selection module 400 determines the grayscale interval with the maximum grayscale sum, and marks the grayscale value corresponding to the center point of the grayscale interval as max;

[0163] The node vector construction module 500 divides the grayscale value range based on the dividing point max, and establishes p and q grayscale sub-intervals in the grayscale interval on both sides after the division, so as to construct a node vector;

[0164] The function fitting segmentation module 600 constructs a B-spline function least squares fitting curve based on the node vector, determines the peak with the highest grayscale value of the least squares fitting curve, and sets the grayscale value of the first trough after the peak as the image segmentation threshold to obtain a hot spot segmentation map.

[0165] The feature extraction module 700 extracts geometric feature parameters of each hot spot target area through the hot spot segmentation map, and the geometric feature parameters include area feature, perimeter feature and shape factor feature.

[0166] The classification model construction module 800 constructs data samples according to geometric feature parameters, divides the data into training sets and test sets, and trains and tests the decision classifier.

[0167] The abnormality determination module 900 identifies the hot spot based on the trained decision classifier and determines the fault type.

[0168] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When executed, the program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media used in the various embodiments provided herein may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0169] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the disclosure in the specification and examples. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered merely as exemplary, and the true scope and spirit of the present disclosure are indicated by the claims.

[0170] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A hot spot detection method based on adaptive B-spline function curve fitting, characterized in that: Contains steps: Collect infrared thermal images of photovoltaic modules under different environmental conditions; Preprocessing the infrared thermal image to obtain a grayscale image, and performing statistics on the grayscale image to establish a grayscale histogram; Establishing n grayscale intervals that evenly divide the grayscale value range, and calculating the grayscale sum of each grayscale interval of the infrared thermal image; Determine the grayscale interval with the maximum grayscale sum, and mark the grayscale value corresponding to the center point of the grayscale interval as max; The grayscale value range is divided based on the dividing point max, and p and q grayscale sub-intervals are established in the grayscale interval on both sides after the division to construct the node vector; Construct a B-spline function least squares fitting curve based on the node vectors, determine the peak with the highest grayscale value of the least squares fitting curve, and set the grayscale value of the first trough after the peak as the image segmentation threshold to obtain a hot spot segmentation map, and mark the hot spot image to obtain a label matrix N; Extracting geometric characteristic parameters of each hot spot target area through the hot spot segmentation map, wherein the geometric characteristic parameters include area characteristics, perimeter characteristics and shape factor characteristics; Construct data samples based on geometric feature parameters, divide them into training sets and test sets, and train and test the decision classifier; Identify hot spots based on the trained decision classifier and determine the fault type; The step of segmenting the grayscale value range based on the dividing point max, and establishing p and q grayscale sub-intervals in the grayscale interval on both sides after segmentation to construct a node vector; The total grayscale value interval is [0, 256]. The grayscale interval [0, max] on the left side of the dividing point is divided into p grayscale subintervals, and the grayscale interval [max, 256] on the right side of the dividing point is divided into q grayscale subintervals. The node vector is: [0,0,0,max / p,2max / p,…,(p-1)max / p,max,max+(255-max) / q,…,max+(q-1)(255-max) / q,255,255,255]; In the step of extracting the geometric characteristic parameters of each hot spot target area through the hot spot segmentation map, in the step of calculating the shape factor characteristics of the hot spot target area image, the formula for obtaining the shape factor characteristics is: ； in, is the shape factor feature of the nth hot spot target area image, is the area feature of the nth hot spot target area image, and G(n) is the perimeter feature of the nth hot spot target area image.

2. The hot spot detection method based on adaptive B-spline function curve fitting according to claim 1, characterized in that: In the step of pre-processing the infrared thermal image to obtain a grayscale image, a Gaussian filter is selected to perform filtering pre-processing on the infrared image of the photovoltaic module; In the step of performing statistics on the grayscale image and establishing a grayscale histogram, the grayscale values ​​of the image pixels are marked based on the preprocessed grayscale image, and the pixel statistics of the grayscale image are performed based on the grayscale values ​​to establish a grayscale histogram. The statistical equation is: ； in, Represents the grayscale value in the grayscale image, Indicates the total number of pixels corresponding to each grayscale value.

3. The hot spot detection method based on adaptive B-spline function curve fitting according to claim 2, characterized in that: In the step of marking the grayscale values ​​of image pixels based on the preprocessed grayscale image, a scanning marking method is adopted in which the grayscale image is scanned column by column and pixel by pixel from left to right and from top to bottom to determine its grayscale value and mark the grayscale value.

4. The hot spot detection method based on adaptive B-spline function curve fitting according to claim 3, characterized in that: In the step of constructing a B-spline function least square fitting curve according to the node vector, the 256 sample points of the grayscale histogram are fitted by the adaptive B-spline function least square method. Fitting is performed, and the curve function after fitting is: ； The selection of the threshold depends on the curve fitting accuracy of the [max, 256] grayscale interval, so the q value is set larger than the p value.

5. The hot spot detection method based on adaptive B-spline function curve fitting according to claim 4, characterized in that: In the step of extracting the geometric feature parameters of each hot spot target area through the hot spot segmentation map, the step of calculating the area feature of the hot spot area is: traversing each hot spot target area image, counting the total number of pixels contained in each hot spot target area image, and obtaining the area feature.

6. The hot spot detection method using adaptive B-spline function curve fitting according to claim 5, characterized in that: In the step of extracting the geometric feature parameters of each hot spot target area through the hot spot segmentation map, the step of calculating the perimeter feature of the hot spot area is: traversing the pixels of the hot spot area image in the hot spot segmentation map, counting the number of all pixels located at the edge of the hot spot area, and obtaining the perimeter feature of the hot spot.

7. The hot spot detection method based on adaptive B-spline function curve fitting according to claim 6, characterized in that: The steps of training the decision classifier specifically include: calculating the optimal division point of each continuous attribute, determining the optimal sample set according to the optimal division point, calculating the information gain corresponding to the optimal sample set, calculating the information gain rate corresponding to each continuous attribute, selecting the optimal division attribute according to the information gain rate, and generating a decision classifier.

8. The hot spot detection method using adaptive B-spline function curve fitting according to claim 1, characterized in that: The system for implementing the hot spot detection method using the adaptive B-spline function curve fitting comprises: Image acquisition module, used to collect infrared thermal images of photovoltaic modules under different environmental conditions; A grayscale processing module pre-processes the infrared thermal image to obtain a grayscale image, and performs statistics on the grayscale image to establish a grayscale histogram; An interval calculation module is configured to establish n grayscale intervals that evenly divide the grayscale value range, and calculate the grayscale sum of each grayscale interval of the infrared thermal image; A marking selection module determines the grayscale interval with the largest grayscale sum, and marks the grayscale value corresponding to the center point of the grayscale interval as max; The node vector construction module divides the grayscale value range based on the dividing point max, and establishes p and q grayscale sub-intervals in the grayscale interval on both sides of the segmentation to construct the node vector; The function fitting segmentation module constructs a B-spline function least squares fitting curve based on the node vector, determines the peak with the highest grayscale value of the least squares fitting curve, and sets the grayscale value of the first trough after the peak as the image segmentation threshold to obtain a hot spot segmentation map; A feature extraction module extracts geometric feature parameters of each hot spot target area through the hot spot segmentation map, wherein the geometric feature parameters include area feature, perimeter feature and shape factor feature; The classification model construction module constructs data samples based on geometric feature parameters, divides the data into training sets and test sets, and trains and tests the decision classifier; The abnormality judgment module identifies hot spots based on the trained decision classifier and determines the fault type.

Images