Multi-dimensional fusion progressive testing method of electroluminescent defect detector
Through a custom spatial resolution test board and a multi-dimensional fusion progressive test method, combining human eye recognition and neural network model, the HYRes algorithm is used to solve the problem of poor repetition and reproducibility of spatial resolution test of electroluminescent defect detectors. , improves measurement accuracy.
Patent Information
- Application Number
- CN202411084027.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-08
- Publication Date
- 2024-12-10
AI Technical Summary
The existing spatial resolution testing methods of electroluminescent defect detectors have problems with poor reproducibility and reproducibility. The human eye discrimination results are inconsistent, and the neural network model may have errors and are limited by the training data range, resulting in low measurement accuracy.
Using a custom multi-dimensional fusion progressive testing method, including self-made spatial resolution test board, grayscale, binarization, edge detection, Hoff detection, neural network model determination and HYRes algorithm, combining human eye recognition and weighted calculation , gradually determine the spatial resolution of rectangular and wedge line pairs.
It effectively reduces human operation errors and model errors, improves the accuracy and repeatability of spatial resolution measurements, and enhances detection accuracy.
Smart Images
Figure CN119107286A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electroluminescent defect detector testing, and in particular to a multi-dimensional fusion and progressive spatial resolution testing method applied to an electroluminescent defect detector. Background Art
[0002] The national metrology technical specification "JJF 2063-2023 Calibration Specification for Electroluminescent Defect Detector for Photovoltaic Cell Modules" proposes the calibration method for electroluminescent defect detectors for photovoltaic modules (hereinafter referred to as EL equipment). Figure 1 As shown in the figure, the basic principle of the electroluminescent defect detector is to pass a forward current through the photovoltaic module to make it emit near-infrared light, and then use a CCD camera to capture the light emitted by the photovoltaic module and form an image. Since the light intensity emitted by the defective part of the photovoltaic module is different from that of the non-defective part, defects in the photovoltaic module that cannot be detected by the human eye can be identified based on the imaging.
[0003] For the imaging system of the electroluminescent defect detector, the national metrology technical specifications use a spatial resolution test plate for calibration. Figure 2 As shown, a stainless steel sheet is etched with multiple groups of hollow patterns on it. One group of hollow patterns consists of 6 equal-width rectangles arranged at equal intervals (the interval is consistent with the width), and one group of stripes corresponds to one spatial resolution (spatial resolution = 0.5 / line pair width). During calibration, the device is pasted on the surface of the photovoltaic module to be tested. After the photovoltaic module is powered on by the electroluminescent defect detector, the electroluminescence of the photovoltaic module passes through the hollow rectangles in the spatial resolution test board to form a line pair group of light and dark. The line pair group is judged by the human eye (from sparse to dense). If 5 black lines can be clearly distinguished, the line pair group passes the judgment. If a line pair group that cannot pass the judgment is found, the spatial resolution corresponding to the previous line pair group of the line pair group is the spatial resolution calibration result of the electroluminescent defect detector imaging system at the corresponding position.
[0004] In the prior art, the spatial resolution of the electroluminescent defect detector is usually determined by human eye judgment. This method may have problems of poor repeatability and reproducibility: that is, for the same line pair group image, different people may have different judgment results, and even if the same person judges at different times, different judgment results may appear. In the prior art, neural networks are also used for recognition. Multiple people judge and vote to determine whether an image passes the judgment. Then, the labeled images are used for training to obtain a trained neural network to judge the image. This method directly fits the average value of multiple people's judgments. The model itself may have certain errors, and due to the limitation of the sampling range of training data, it is easy to further increase the error of the judgment result due to the insufficient sampling range of training data.
[0005] In view of this, for the calibration method of the EL device imaging system in the "JJF 2063-2023 Photovoltaic Cell Module Electroluminescent Defect Detector Calibration Specification", for the spatial resolution test method, it is necessary to find a line pair discrimination method with high accuracy and high repeatability to improve the spatial resolution measurement accuracy. Summary of the invention
[0006] The technical problem to be solved by the present invention is to provide a multi-dimensional fusion progressive testing method for an electroluminescent defect detector to reduce the impact of human operating errors, human eye misjudgment and model sample defects on the spatial resolution test of the electroluminescent defect detector.
[0007] The problem of the present invention is achieved as follows: A multi-dimensional fusion progressive testing method for an electroluminescent defect detector comprises the following steps:
[0008] Step S1, customizing a spatial resolution test plate, wherein the spatial resolution test plate includes a hollow positioning circle, a tilt angle detection bar, a rectangular line pair group and a wedge-shaped line pair group, wherein the rectangular line pair group and the wedge-shaped line pair group are both provided with line pairs including horizontal direction, vertical direction, 45° tilt direction and 135° tilt direction;
[0009] Step S2, using the spatial resolution test plate to perform a calibration operation on the electroluminescent defect detector to obtain corresponding image information;
[0010] Step S3, performing the following operations on the obtained image: grayscale - binarization - edge detection - Hough detection, obtaining the image tilt angle, and correcting the image angle and area position;
[0011] Step S4, automatically intercepting all rectangular line pair groups and wedge line pair groups according to the distance relationship between each line pair position and the center of the positioning circle, and for each rectangular line pair image, rotating the rectangular line pair images in the horizontal direction, the 45° inclined direction, and the 135° inclined direction according to its angle, so that the long side of the rectangle is parallel to the vertical direction;
[0012] Step S5, constructing a neural network model to judge the rectangular line pair image, first extracting features from the rectangular line pair image, and using a spatial pyramid pooling layer to output a uniform size, and then obtaining a corresponding line pair image judgment value J1 through a classifier;
[0013] Step S6, obtaining the rectangular line pair image in step S4, averaging the image grayscale along the long side direction of the black and white line pair, obtaining a grayscale curve with the image width as the independent variable and the grayscale mean as the dependent variable, and using the SR02 black line detection algorithm in the HYRes algorithm to obtain the number of black lines in the grayscale curve. If the number of black lines is not 5, the output judgment value J2 is 0, otherwise, the output judgment value J2 is 1;
[0014] Step S7, obtaining the rectangular line pair image in step S4 and enlarging it, judging the rectangular line pair image by human eyes, and when it can be identified that the number of black lines is 5, the judgment value J3 is 1, otherwise, J3 is 0;
[0015] Step S8, calculate the final discrimination value J, J = a × J1 + b × J2 + c × J3, where a, b, c are weight values of the three discrimination methods, and satisfy a + b + c = 1, if J is greater than or equal to 0.5, then the rectangular line pair image discrimination result is discriminable, otherwise it is undiscriminable;
[0016] Step S9, repeating steps S5 to S8, judging one by one from the rectangular line pair image corresponding to the low spatial resolution to the rectangular line pair image corresponding to the high spatial resolution, if the judgment result of a rectangular line pair image is judged to be judged, then continue to judge the next rectangular line pair image, until a rectangular line pair image is found to be unjudgable, and record the spatial resolution corresponding to the previous rectangular line pair image with a judged result of being judged as the initial spatial resolution R of the EL device at the corresponding position and direction s1 ;
[0017] Step S10: Find the spatial resolution R corresponding to the last distinguishable line pair in the rectangular line pair. s1 The spatial resolution R corresponding to the first unresolvable line pair s2 , combined with the wedge line pair images in the same direction, analyze: first rotate the wedge line pair to the vertical direction, and find the image with a spatial resolution of R in the wedge line pair image s1 and R s2 The corresponding row, from R s1 The corresponding row starts at R s2 The number of black lines is measured row by row until a row where the number of black lines is not 5 is found. The corresponding spatial resolution is the final spatial resolution Rs of the EL device at the corresponding position and direction.
[0018] Furthermore, the calibration process in step S2 is specifically as follows: the spatial resolution test board is adhered to the surface of the photovoltaic module to be tested, and after the photovoltaic module is energized by the electroluminescent defect detector, the electroluminescence of the photovoltaic module is transmitted through the positioning circle, tilt angle detection bar, rectangular line pair group and wedge line pair group in the spatial resolution test board to obtain corresponding image information.
[0019] Furthermore, the step S3 specifically includes: establishing a Cartesian coordinate system, with the geometric center of the image as the zero point, the horizontal right direction as the x-axis square, and the vertical upward direction as the y-axis positive direction;
[0020] Perform the following operations on the obtained image: grayscale - binarization - edge detection - Hough detection to obtain the center coordinates (x1, y1) of the positioning circle, the coordinates (x2, y2), (x3, y3) of the two endpoints of the inclination angle detection bar, and its included angle θ with the y-axis. Rotate the image by -θ so that the inclination angle detection bar remains vertical.
[0021] After that, continuously perform the following operations on the image: rotate the image by 90° with the image center as the rotation center, or flip it along the x-axis and y-axis until the following conditions are met simultaneously: x1 < x2 and x1 < x3; y1 > 0.5×(y2 + y3).
[0022] Further, the specific steps of step S5 include: constructing a neural network model to determine the image: first, extract features of the image through multiple feature modules. Each feature extraction module includes a convolutional layer, a ReLU activation function layer, and a max pooling layer. The convolutional layer is used to extract features in each dimension of the image. The ReLU activation function layer is used to introduce non-linearity to improve the expression ability of the model. The max pooling layer is used to reduce the image size dimension. After the last feature extraction module is executed, a spatial pyramid pooling layer is used to unify the sizes of image features of different sizes and then output.
[0023] After that, connect a classifier composed of a certain number of fully connected layers to output the determination value J1 of the rectangular line pair image, where 0 <= J1 <= 1.
[0024] Further, step S5 also includes: collecting line pair images taken under different EL devices, different shooting parameters, and different photovoltaic modules. For each line pair image, select N people to judge M times, where both N and M are odd numbers. For each person and each line pair image, take the majority result of the M judgments as the discrimination result of this person for this image. For each image, take the majority result of the N people as the final discrimination result of this image. Obtain a set of line pair images with labels through the above method for training the neural network model.
[0025] Further, "obtaining the number of black lines in the grayscale curve by using the SR02 black line detection algorithm in the HYRes algorithm" in step S6 specifically refers to:
[0026] For a grayscale curve graph, determine the number of black lines in the image by finding the number of minimum values in the curve. At the same time, set a threshold. Only when the increase of the local minimum value exceeds the threshold, it is determined that there is a black line in the image. And as the line pair density increases, the threshold gradually decreases to adapt to the decrease of the grayscale amplitude.
[0027] Furthermore, in step S7, for a black line in a line pair image, when it is identifiable that the length of the black line accounts for more than 50% of the total length of the line pair image, the black line is defined to be identifiable.
[0028] Furthermore, a, b, and c in step S8 are all set to 1 / 3, that is,
[0029]
[0030] Furthermore, the step S10 is specifically as follows:
[0031] Find the spatial resolution R corresponding to the last distinguishable line pair in the rectangular line pair s1 The spatial resolution R corresponding to the first unresolvable line pair s2 , analyze the image by combining wedge lines in the same direction:
[0032] If the wedge line pair is in the horizontal direction, inclined at 45° or inclined at 135°, it is rotated so that the wedge line pair image is arranged vertically, that is, the low spatial resolution area is at the bottom of the screen and the high spatial resolution area is at the top of the screen;
[0033] Find the spatial resolution R in the wedge line pair image s1 and R s2 The corresponding row, from R s1 The corresponding row starts at R s2 The corresponding rows use the SR02 black line detection algorithm in the HYRes algorithm to measure the number of black lines row by row until a row where the number of black lines is not 5 is found. The corresponding spatial resolution is the final spatial resolution Rs of the EL device at that position and in that direction.
[0034] When the SR02 black line detection algorithm in the HYRes algorithm is used to measure the number of black lines line by line, the grayscale of the image is averaged along the long side of the black and white line pairs in advance to obtain a grayscale curve with the image width as the independent variable and the grayscale mean as the dependent variable. The SR02 black line detection algorithm in the HYRes algorithm is then used to obtain the number of black lines in the grayscale curve. For each grayscale curve, the number of minimum values in the curve is found to determine the number of black lines in the image. A threshold is set. Only when the increase in the local minimum value exceeds the threshold, the image is judged to have black lines. As the line pair density increases, the threshold is gradually reduced to adapt to the decrease in grayscale amplitude.
[0035] The advantages of the present invention are: through a self-made high-precision spatial resolution test board, combined with human eye recognition, a neural network model and a detection algorithm to weightedly calculate the corresponding spatial resolution, and through wedge line pairs to further find a higher precision spatial resolution, the problem of poor repeatability caused by directly using human eye judgment can be effectively avoided, while improving the measurement accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.
[0037] Figure 1 Schematic diagram of the test principle of the electroluminescent defect detector.
[0038] Figure 2 This is a schematic diagram of the structure of a spatial resolution test board used in existing electroluminescent defect detectors.
[0039] Figure 3 A schematic diagram of the structure of a spatial resolution test board used in a multi-dimensional fusion progressive test method of an electroluminescent defect detector of the present invention.
[0040] Figure 4 The present invention is a flowchart of the execution of a multi-dimensional fusion progressive testing method for an electroluminescent defect detector. DETAILED DESCRIPTION
[0041] The overall design idea of the present invention is as follows: a self-made improved spatial resolution test board is used, and for each rectangular line pair, discrimination values J1, J2, and J3 are obtained through improved neural network fitting, grayscale curve analysis, and human eye discrimination in three dimensions, and the discrimination value J of the rectangular pair is obtained in a weighted manner, and judgments are made one by one from sparse line pairs to dense line pairs, and the spatial resolution corresponding to the last discriminable line pair is found as the initial value of the spatial resolution, and then combined with the wedge line pair, the final value of the spatial resolution is calculated, and a discrimination result with higher accuracy and smaller error can be obtained.
[0042] In order to make the present invention more clearly understood, a preferred embodiment is now described in detail with reference to the accompanying drawings.
[0043] like Figure 4 As shown, a multi-dimensional fusion progressive testing method for an electroluminescent defect detector includes the following steps:
[0044] Step S1, customizing a spatial resolution test plate, the spatial resolution test plate including a hollow positioning circle, a tilt angle detection bar, a rectangular line pair group and a wedge-shaped line pair group, wherein the rectangular line pair group and the wedge-shaped line pair group are both provided with line pairs including horizontal direction, vertical direction, 45° tilt direction and 135° tilt direction; Figure 3As shown in the figure, the spatial resolution test board of the present invention includes a positioning area 1, an inclination angle detection area 2, a rectangular line pair area 3, and a wedge-shaped line pair area 4. The hollow positioning circle is located within the positioning area 1, and the inclination angle detection strip is located within the inclination angle detection area 2 and is arranged in the vertical direction. The rectangular line pair group is arranged within the rectangular line pair area 3, and a plurality of rectangular line pairs corresponding to different resolutions are arranged in each direction. The wedge-shaped line pair group is arranged within the wedge-shaped line pair area 4, and for each rectangular line pair, a wedge-shaped line pair with the same direction can be found.
[0045] Step S2: Use the spatial resolution test board to perform a calibration operation on the electroluminescence defect detector to obtain corresponding image information. Preferably, the calibration process in step S2 is specifically as follows: Affix the spatial resolution test board to the surface of the photovoltaic module to be tested. After powering on the photovoltaic module through the electroluminescence defect detector, the electroluminescence of the photovoltaic module passes through the positioning circle, inclination angle detection strip, rectangular line pair group, and wedge-shaped line pair group in the spatial resolution test board to obtain corresponding image information.
[0046] Step S3: Establish a Cartesian coordinate system with the geometric center of the image as the origin, the horizontal right direction as the positive x-axis, and the vertical upward direction as the positive y-axis. Perform the following operations on the obtained image: grayscale - binarization - edge detection - Hough detection to obtain the center coordinates (x1, y1) of the positioning circle, the coordinates of the two endpoints (x2, y2), (x3, y3) of the inclination angle detection strip, and its included angle θ with the y-axis. Rotate the image by -θ to keep the inclination angle detection vertical. Then, continuously perform the following operations on the image: Rotate the image by 90° around the image center, or flip it along the x-axis and y-axis until the following two conditions are simultaneously met: (1) x1 < x2 and x1 < x3; (2) y1 > 0.5×(y2 + y3);
[0047] Step S4: According to the distance relationship between the positions of each line pair and the center (x1, y1) of the positioning circle, use a computer to automatically intercept all the rectangular line pair areas and wedge-shaped line pair areas. For each line pair image, rotate the rectangular line pair images in the horizontal, 45° inclined, and 135° inclined directions according to their inclination angles, and finally make the long sides of the rectangles parallel to the vertical direction;
[0048] Step S5: Construct a neural network model to determine the rectangular line pair images. First, extract features from the rectangular line pair images, and use the spatial pyramid pooling layer to output a unified size, and then obtain the determination value J1 of the corresponding line pair image through a classifier;
[0049] Step S6, obtaining the rectangular line pair image in step S4, averaging the image grayscale along the long side direction of the black and white line pair, obtaining a grayscale curve with the image width (pixel point) as the independent variable and the grayscale mean as the dependent variable, and using the SR02 black line detection algorithm in the HYRes algorithm to obtain the number of black lines in the grayscale curve. If the number of black lines is not 5, the output judgment value J2 is 0, otherwise, the output judgment value J2 is 1;
[0050] Step S7, obtain the rectangular line pair image in step S3 and enlarge it, the magnification rate is set according to actual needs, and the rectangular line pair image is judged by human eyes. When the number of black lines that can be identified is 5, the judgment value J3 is 1, otherwise, J3 is 0; preferably, for a black line in a line pair image, when it can be identified that the length of the black line accounts for more than 50% of the total length of the line pair image, it is defined that the black line can be identified.
[0051] Step S8, calculate the final discrimination value J, J = a × J1 + b × J2 + c × J3, where a, b, c are weight values of the three discrimination methods, and satisfy a + b + c = 1, if J is greater than or equal to 0.5, then the rectangular line pair image discrimination result is discriminable, otherwise it is undiscriminable; preferably, a, b, c in step S8 are set according to actual conditions, such as they can be set to 1 / 3 according to actual conditions, that is,
[0052]
[0053] Step S9, repeating steps S5 to S8, judging one by one from the rectangular line pair image corresponding to the low spatial resolution to the rectangular line pair image corresponding to the high spatial resolution, if the judgment result of a rectangular line pair image is judged to be judged, then continue to judge the next rectangular line pair image, until a rectangular line pair image is found to be unjudgable, and record the spatial resolution corresponding to the previous rectangular line pair image with a judged result of being judged as the initial spatial resolution R of the EL device at this position and direction s1 ;
[0054] Step S10: Find the spatial resolution R corresponding to the last distinguishable line pair in the rectangular line pair. s1 The spatial resolution R corresponding to the first unresolvable line pair s2 , combined with the wedge line pair images in the same direction, analyze: first rotate the wedge line pair to the vertical direction, and find the image with a spatial resolution of R in the wedge line pair image s1 and R s2 The corresponding row, from R s1 The corresponding row starts at R s2 The number of black lines is measured row by row until a row where the number of black lines is not 5 is found. The corresponding spatial resolution is the final spatial resolution Rs of the EL device at the corresponding position and direction.
[0055] Preferably, the step S5 specifically includes: constructing a neural network model to judge the image: first, the image is subjected to feature extraction through multiple feature modules (each feature module has the same structure and different parameters), each feature extraction module includes a convolution layer, a ReLU activation function layer and a maximum pooling layer, wherein the convolution layer is used to extract features of each dimension of the image, the ReLU activation function layer is used to introduce nonlinearity to improve the expression ability of the model, and the maximum pooling layer is used to reduce the image size dimension. After executing the last feature extraction module, the image features of different sizes are unified in size through a spatial pyramid pooling layer and then output;
[0056] Then, a classifier consisting of a certain number of fully connected layers is connected to output the rectangular line pair image judgment value J1, 0<=J1<=1.
[0057] Preferably, step S5 also includes: preparing a training set in the following manner: pasting a spatial resolution test board on different types of photovoltaic modules (different numbers of grid lines, different technology types (PERC, topcon, HJT, etc.)), using EL devices produced by different manufacturers, using different shooting parameter settings (parameters include: exposure time, ISO, power-on current, etc.), shooting the spatial resolution test board image, separating all rectangular line pair images therein, and rotating them to a vertical direction, copying all line pair images N copies (odd number of copies) and shuffling the order, then selecting M (odd number of) judges, each judger judges all line pair images, and each person judges each image N times in total. For each judger, the judgment result of each image is based on the majority of the N results. For each image, the final judgment result is based on the majority of the M personnel. Finally, a certain number of line pair images with "judgment passed" or "judgment failed" labels are obtained, and the neural network model is trained using this as the training set of the neural network.
[0058] Preferably, in step S6, "using the SR02 black line detection algorithm in the HYRes algorithm to obtain the number of black lines in the grayscale curve" is specifically as follows: for a grayscale curve graph, the number of black lines in the image is determined by finding the number of minimum values in the curve, and at the same time, in order to avoid the influence of image noise interference on the result, a threshold is set, and only when the increase of the local minimum value exceeds the threshold, it is determined that there are black lines in the image, and as the line pair density increases, the threshold is gradually reduced to adapt to the decrease in grayscale amplitude.
[0059] Preferably, the step S10 is specifically as follows:
[0060] Find the spatial resolution R corresponding to the last distinguishable line pair in the rectangular line pair s1The spatial resolution R corresponding to the first unresolvable line pair s2 , analyze the image by combining wedge lines in the same direction:
[0061] If the wedge line pair is in the horizontal direction, inclined at 45° or inclined at 135°, it is rotated so that the wedge line pair image is arranged vertically, that is, the low spatial resolution area is at the bottom of the screen and the high spatial resolution area is at the top of the screen;
[0062] Find the spatial resolution R in the wedge line pair image s1 and R s2 The corresponding row, from R s1 The corresponding row starts at R s2 The corresponding rows use the SR02 black line detection algorithm in the HYRes algorithm to measure the number of black lines row by row until a row where the number of black lines is not 5 is found. The corresponding spatial resolution is the final spatial resolution Rs of the EL device at that position and in that direction.
[0063] When the SR02 black line detection algorithm in the HYRes algorithm is used to measure the number of black lines line by line, the grayscale of the image is averaged along the long side of the black and white line pairs in advance to obtain a grayscale curve with the image width as the independent variable and the grayscale mean as the dependent variable. The SR02 black line detection algorithm in the HYRes algorithm is then used to obtain the number of black lines in the grayscale curve. For each grayscale curve, the number of minimum values in the curve is found to determine the number of black lines in the image. A threshold is set. Only when the increase in the local minimum value exceeds the threshold, the image is judged to have black lines. As the line pair density increases, the threshold is gradually reduced to adapt to the decrease in grayscale amplitude.
[0064] The present invention adopts the existing commonly used test method for the spatial resolution of the electroluminescent defect detector (human eye discrimination and neural network model), and introduces the HYRes black line detection algorithm in the "CPIA DC-003 Digital Camera Resolution Measurement Method" standard to discriminate the spatial resolution of the electroluminescent defect detector of the present invention, and obtains the discrimination result of the rectangular line pair after weighting the three discrimination methods. Since the spatial resolution of the rectangular line pair is set relatively discretely and the accuracy is low, in order to further improve the test accuracy, the present invention uses a self-made test board, adds a wedge line pair to the test board, and sets a positioning area and an angle detection area to realize automatic discrimination, reduce operation errors, and improve detection accuracy.
[0065] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.
Claims
1. A multi-dimensional fusion progressive testing method for an electroluminescent defect detector, characterized in that: The steps include: Step S1, customizing a spatial resolution test plate, wherein the spatial resolution test plate includes a hollow positioning circle, a tilt angle detection bar, a rectangular line pair group and a wedge-shaped line pair group, wherein the rectangular line pair group and the wedge-shaped line pair group are both provided with line pairs including horizontal direction, vertical direction, 45° tilt direction and 135° tilt direction; Step S2, using the spatial resolution test plate to perform a calibration operation on the electroluminescent defect detector to obtain corresponding image information; Step S3, performing the following operations on the obtained image: grayscale - binarization - edge detection - Hough detection, obtaining the image tilt angle, and correcting the image angle and area position; Step S4, automatically intercepting all rectangular line pair groups and wedge line pair groups according to the distance relationship between each line pair position and the center of the positioning circle, and for each rectangular line pair image, rotating the rectangular line pair images in the horizontal direction, the 45° inclined direction, and the 135° inclined direction according to its angle, so that the long side of the rectangle is parallel to the vertical direction; Step S5, constructing a neural network model to judge the rectangular line pair image, first extracting features from the rectangular line pair image, and using a spatial pyramid pooling layer to output a uniform size, and then obtaining a corresponding line pair image judgment value J1 through a classifier; Step S6, obtaining the rectangular line pair image in step S4, averaging the image grayscale along the long side direction of the black and white line pair, obtaining a grayscale curve with the image width as the independent variable and the grayscale mean as the dependent variable, and using the SR02 black line detection algorithm in the HYRes algorithm to obtain the number of black lines in the grayscale curve. If the number of black lines is not 5, the output judgment value J2 is 0, otherwise, the output judgment value J2 is 1; Step S7, obtaining the rectangular line pair image in step S4 and enlarging it, judging the rectangular line pair image by human eyes, and when it can be identified that the number of black lines is 5, the judgment value J3 is 1, otherwise, J3 is 0; Step S8, calculate the final discrimination value J, J = a × J1 + b × J2 + c × J3, where a, b, c are weight values of the three discrimination methods, and satisfy a + b + c = 1, if J is greater than or equal to 0.5, then the rectangular line pair image discrimination result is discriminable, otherwise it is undiscriminable; Step S9, repeating steps S5 to S8, judging one by one from the rectangular line pair image corresponding to the low spatial resolution to the rectangular line pair image corresponding to the high spatial resolution, if the judgment result of a rectangular line pair image is judged to be judged, then continue to judge the next rectangular line pair image, until a rectangular line pair image is found to be unjudgable, and record the spatial resolution corresponding to the previous rectangular line pair image with a judged result of being judged as the initial spatial resolution R of the EL device at the corresponding position and direction s1 ; Step S10: Find the spatial resolution R corresponding to the last distinguishable line pair in the rectangular line pair. s1 The spatial resolution R corresponding to the first unresolvable line pair s2 , combined with the wedge line pair images in the same direction, analyze: first rotate the wedge line pair to the vertical direction, and find the image with a spatial resolution of R in the wedge line pair image s1 and R s2 The corresponding row, from R s1 The corresponding row starts at R s2 The number of black lines is measured row by row until a row where the number of black lines is not 5 is found. The corresponding spatial resolution is the final spatial resolution Rs of the EL device at the corresponding position and direction.
2. A multi-dimensional fusion progressive testing method for an electroluminescent defect detector according to claim 1, characterized in that: The calibration process in step S2 is specifically as follows: the spatial resolution test board is pasted to the surface of the photovoltaic module to be tested, and after the photovoltaic module is powered on by the electroluminescent defect detector, the electroluminescence of the photovoltaic module is transmitted through the positioning circle, tilt angle detection bar, rectangular line pair group and wedge line pair group in the spatial resolution test board to obtain corresponding image information.
3. The multi-dimensional fusion progressive testing method of an electroluminescent defect detector according to claim 1, characterized in that: The step S3 specifically includes: establishing a Cartesian coordinate system, with the geometric center of the image as the zero point, the horizontal right direction as the x-axis square, and the vertical upward direction as the y-axis positive direction; The obtained image is subjected to the following operations: grayscale-binarization-edge detection-Hough detection, and the coordinates of the center of the positioning circle (x1, y1), the coordinates of the two end points of the tilt angle detection bar (x2, y2), (x3, y3) and the angle θ between them and the y-axis are obtained, and the image is rotated by -θ so that the tilt angle detection bar remains vertical; After that, the following operations are continuously performed on the image: rotating the image by 90° with the center of the image as the rotation center, or flipping the image along the x-axis and y-axis, until the following conditions are satisfied simultaneously: x1 < x2 and x1 < x3; y1 > 0.5×(y2 + y3).
4. The multi-dimensional fusion progressive testing method of an electroluminescent defect detector according to claim 1, characterized in that: The specific steps of step S5 include: constructing a neural network model to judge the image: first, extracting features of the image through multiple feature modules. Each feature extraction module includes a convolutional layer, a ReLU activation function layer, and a max pooling layer. The convolutional layer is used to extract features of each dimension of the image, the ReLU activation function layer is used to introduce non-linearity to improve the expression ability of the model, and the max pooling layer is used to reduce the size dimension of the image. After the last feature extraction module is executed, a spatial pyramid pooling layer is used to unify the sizes of image features with different sizes and then output; Then, a classifier composed of a certain number of fully connected layers is connected to output the determination value J1 of the rectangular line pair image, where 0 <= J1 <= 1.
5. The multi-dimensional fusion progressive testing method of an electroluminescent defect detector according to claim 1, characterized in that: Step S5 further includes: collecting line pair images taken under different EL devices, different shooting parameters, and different photovoltaic modules. For each line pair image, N people are selected to judge M times, where both N and M are odd numbers. For each person and each line pair image, the majority result of the M judgments is taken as the discrimination result of the person for the image. For each image, the majority result of the N people is taken as the final discrimination result of the image. The labeled line pair image set is obtained through the above method and used to train the neural network model.
6. The multi-dimensional fusion progressive testing method of an electroluminescent defect detector according to claim 1, characterized in that: In step S6, "obtaining the number of black lines in the gray curve by using the SR02 black line detection algorithm in the HYRes algorithm" is specifically as follows: For a gray curve graph, the number of black lines in the image is discriminated by finding the number of minimum values in the curve. At the same time, a threshold is set. Only when the increase in the local minimum value exceeds the threshold, it is discriminated that there is a black line in the image. And as the line pair density increases, the threshold gradually decreases to adapt to the decrease in the gray amplitude.
7. The multi-dimensional fusion progressive testing method of an electroluminescent defect detector according to claim 1, characterized in that: In step S7, for a black line in a line pair image, when the proportion of the length of the recognized black line in the total length of the line pair image exceeds 50%, it is defined that the black line can be recognized.
8. The multi-dimensional fusion progressive testing method of an electroluminescent defect detector according to claim 1, characterized in that: In step S8, a, b, and c are all set to 1 / 3, and then 9. The multi-dimensional fusion progressive testing method of an electroluminescent defect detector according to claim 1, characterized in that: The specific steps of step S10 are as follows: Find the spatial resolution R corresponding to the last distinguishable line pair in the rectangular line pair s1 The spatial resolution R corresponding to the first unresolvable line pair s2 , analyze the image by combining wedge lines in the same direction: If the wedge line pair is in the horizontal direction, the 45° inclined direction, or the 135° inclined direction, then rotate it so that the wedge line pair image is vertically arranged, that is, the low spatial resolution area is below the screen and the high spatial resolution area is above the screen; Find the spatial resolution R in the wedge line pair image s1 and R s2 The corresponding row, from R s1 The corresponding row starts at R s2 The corresponding rows use the SR02 black line detection algorithm in the HYRes algorithm to measure the number of black lines row by row until a row where the number of black lines is not 5 is found. The corresponding spatial resolution is the final spatial resolution Rs of the EL device at that position and in that direction. When the SR02 black line detection algorithm in the HYRes algorithm is used to measure the number of black lines line by line, the grayscale of the image is averaged along the long side of the black and white line pairs in advance to obtain a grayscale curve with the image width as the independent variable and the grayscale mean as the dependent variable. The SR02 black line detection algorithm in the HYRes algorithm is then used to obtain the number of black lines in the grayscale curve. For each grayscale curve, the number of minimum values in the curve is found to determine the number of black lines in the image. A threshold is set. Only when the increase in the local minimum value exceeds the threshold, the image is judged to have black lines. As the line pair density increases, the threshold is gradually reduced to adapt to the decrease in grayscale amplitude.