A Method for Detecting the Cross Intersection Points of the Edges of Photovoltaic Modules Based on Infrared Images

Abstract

The present invention discloses a method for detecting the cross intersection points at the edges of photovoltaic modules based on infrared images. By analyzing the imaging characteristics of photovoltaic modules in infrared images and the commonalities and differences existing in various module arrangement scenarios, a method for detecting the cross intersection points at the edges of photovoltaic modules is proposed, which can achieve rapid detection of the cross intersection points formed by the edges of adjacent modules in infrared images and exhibits good generality and accuracy in various scenarios; the detection of the cross intersection points provides assistance for subsequent links of photovoltaic inspection and can effectively reduce the difficulty of links such as module segmentation, module detection, and module positioning.

Description

Technical Field

[0001] The invention belongs to the technical field of unmanned aerial vehicle inspection, and in particular relates to a method for detecting cross intersections of photovoltaic module edges based on infrared images. Background Art

[0002] The safety issues and power generation loss problems of power stations have always been the pain points of concern to owners. Hot spots cause components to continue to heat up, which can seriously burn components and cause fires, leading to losses of power assets and equipment and casualties. The power generation losses caused by photovoltaic components account for more than 70%, and photovoltaic components need to be inspected in a refined manner, accurately operated and maintained, to increase power generation and improve economic benefits.

[0003] The traditional method is to conduct inspection and maintenance manually. Due to the large area of ​​the power station, many equipment, inappropriate location, difficult manual inspection, large workload, low efficiency and high cost, it is difficult to guarantee the inspection coverage and frequency. The emergence of drones can alleviate the pressure of traditional methods. Intelligent drones have high maneuverability and can move freely without being restricted by terrain. They can carry a variety of mission equipment. In the inspection of photovoltaic systems, dual-light cameras can be used to detect the components through infrared images. Abnormal components will show heating phenomena, and they can be automatically detected and located through deep learning AI methods. Due to the wide distribution of photovoltaic panels, effective target positioning methods play a vital role in drone inspections. It can not only realize more accurate digital management of power stations and greatly improve the replacement efficiency of fault elimination personnel, but also open up the integration channel between drone inspections and diagnostic information in other dimensions.

[0004] Photovoltaic scenes can present a variety of forms due to factors such as site topography, component arrangement, and grid-connected scale. This puts higher demands on the detection and positioning of abnormal components. If the components in the image can be segmented, the robustness of the detection and positioning algorithms can be effectively improved. However, due to different shooting environments, light, and component materials, there is a lot of noise in the straight line edge information in the image, and the straight line detection method has many defects and cannot meet the requirements of accurate segmentation.

[0005] There are many difficulties in the image processing stage of drone-based intelligent inspection, which leads to poor versatility of the algorithm in different scenarios. Therefore, it is necessary to develop a detection method that can effectively solve the segmentation problem of components in a variety of complex scenarios. Summary of the invention

[0006] The purpose of the present invention is to overcome the above-mentioned deficiencies of the prior art and provide a method for detecting cross intersections of photovoltaic module edges based on infrared images.

[0007] To solve the technical problem, the technical solution of the present invention is: a method for detecting the cross intersection at the edge of a photovoltaic module based on an infrared image, comprising the following steps:

[0008] Step 1: Obtain infrared images of multiple scenarios by means of a drone, preprocess the infrared images, and remove high-frequency noise in the photovoltaic module area through mean filtering;

[0009] Step 2: Filter the preprocessed infrared image through a custom operator, detect the pixel areas that meet the requirements, and determine the cross intersection candidate areas through Hough transform and morphological processing;

[0010] Step 3: Filter the pixels in the cross intersection candidate areas through a custom operator to detect the cross intersections in the photovoltaic module area.

[0011] Preferably, step 1 is specifically: obtain infrared images of multiple scenarios by shooting with a drone, and the camera on the drone shoots along the direction of the bracket of the photovoltaic module. Set the line with an up-down trend in the infrared image as the meridian line, and the line with a left-right trend as the parallel line. The meridian line is long and the parallel line is short; control the height of the drone relative to the photovoltaic module and the equivalent focal length of the camera to ensure that the edge width of the photovoltaic module is 3 to 30 pixels.

[0012] Preferably, in step 1, high-frequency noise in the photovoltaic module area is removed through mean filtering, and the mean filtering operator for mean filtering is 5*5.

[0013] Preferably, the multiple scenarios include mountain scenarios, water surface scenarios, roof scenarios, four-row plain scenarios, flat single-axis scenarios, and inclined single-axis scenarios.

[0014] Preferably, step 2 is specifically:

[0015] Step 2-1: Filter through a custom operator. The operator is a 1*n matrix, which includes three parts: the current pixel area, the component border area, and the component area. Filter each pixel through the custom operator to detect the pixel point area with the meridian line feature of the photovoltaic module;

[0016] Step 2-2: Determine the cross intersection candidate areas through Hough transform and morphological processing.

[0017] Preferably, the custom operator filtering in step 2-1 is:

[0018] 2-1-1) Take the current pixel as the center, determine the window area range, and judge whether the current pixel value is the minimum value in the window area within this range. If it does not meet the requirement, directly jump out and execute the filtering of the next pixel, and indicate that the current pixel is not on the meridian line with a cross intersection. If it meets the requirement, then execute the next step;

[0019] 2-1-2) Compare the pixel values of each pixel in the component area with the pixels adjacent to itself. If a large difference appears, immediately jump out and perform the next pixel filtering. If all the pixels in the component area meet the requirement of a low difference, it indicates that the current pixel meets the requirement, and the warp area image that meets the requirement is detected.

[0020] Preferably, the range of the window area is 20 to 30 pixels.

[0021] Preferably, in step 2-2, the specific process of determining the cross intersection candidate area by the Hough transform and morphological operations is as follows: perform the Hough transform on the warp area image to realize the straight line fitting of the real warp area. The fitted straight line is redrawn and processed by morphological dilation, and then the continuous cross intersection candidate area is obtained.

[0022] Preferably, step 3 includes the following steps:

[0023] 3-1) Use the cross intersection candidate area image as the mask image, and for each valid pixel in it, find the corresponding pixel position in the preprocessed infrared image;

[0024] 3-2) For this pixel position, determine a square area around it, and respectively determine the upper, lower, left, and right four areas on the four sides of the square area. These four areas are pixel areas of 1*n or n*1;

[0025] 3-3) Find the pixel with the minimum pixel value in each area;

[0026] 3-4) Calculate the position differences Δxt, Δxb, Δyl, and Δyr between the coordinates of the minimum pixels on the four sides and the center point of the square;

[0027] 3-5) If the differences between Δxt and Δxb and between Δyl and Δyr can both be guaranteed to be within the range of 10 pixels, then this pixel point is considered as the cross intersection point.

[0028] Preferably, the square area in step 3-1 is a square.

[0029] Compared with the prior art, the advantages of the present invention are as follows:

[0030] (1) The present invention proposes a method for detecting the cross intersections at the edges of photovoltaic modules based on infrared images. This detection method relies on various real - scene data, combines the influences of multiple factors such as light, noise, and morphology, analyzes the commonalities and differences presented in the images by sorting out the real data of multiple scenarios, and extracts the idea of detecting cross intersections. Since the metal frame at the edge of the photovoltaic module has a different temperature from the battery cells, the infrared images can present relatively stable cross - intersection edges in many scenarios. The detection of this feature can effectively improve the accuracy and stability of component edge segmentation, thereby enhancing the robustness of subsequent anomaly detection algorithms and component positioning algorithms, and reducing the difficulty of subsequent detection algorithms and positioning algorithms;

[0031] (2) The present invention first processes the entire initial infrared image using the mean - filtering algorithm. While ensuring that the relative brightness of the edge region of the photovoltaic module remains unchanged, it eliminates the high - frequency noise in other regions, guarantees the output quality, maximally saves the processing time, and improves the robustness of subsequent steps;

[0032] (3) The present invention uses a custom operator for filtering, as well as Hough transform and morphological processing, which narrows the candidate region of the cross intersections, further reduces the probability of false detection, and improves the detection speed;

[0033] (4) Finally, the present invention directly performs cross - intersection detection on the pre - processed image by filtering each pixel in the candidate region of the cross intersections using a custom operator, and compares the narrowed candidate region of the cross intersections with the pre - processed image, which improves the efficiency while ensuring accuracy, and effectively enhances the robustness of subsequent anomaly detection algorithms and component positioning algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 shows the diverse scenarios applied in the present invention;

[0035] Figure 2 is the schematic diagram of the pre - processing in step 1 of the present invention;

[0036] Figure 3 is the schematic diagram of the custom operator in step 2 of the present invention;

[0037] Figure 4 is the schematic diagram of cross - intersection detection in step 3 of the present invention;

[0038] Figure 5 is the schematic diagram of the processing flow of a method for detecting the cross intersections at the edges of photovoltaic modules based on infrared images according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0039] The following describes the specific embodiments of the present invention in conjunction with the embodiments:

[0040] As Figure 1As shown in the figure, the diverse scenarios to which the present invention is applied. Photovoltaic power stations can be divided into fixed-axis, flat single-axis, inclined single-axis, etc. according to the steering flexibility of the brackets; they can be divided into flat ground scenarios, mountain scenarios, water surface scenarios, rooftops, etc. according to the installation location and terrain; there are also a variety of component arrangement methods within the same bracket or string. Due to the existence of the above differences, there are many difficulties in the image processing stage of intelligent inspection based on drones, resulting in poor generality of the algorithm in different scenarios. The present invention analyzes the commonalities and differences shown in the images by sorting out the real data of multiple scenarios, and extracts the idea of detecting the cross intersection points.

[0041] As Figure 1 shown, because the metal frame of the component edge and the cell temperature are different, in many scenarios (except for a few single rows), the infrared image can present a relatively stable cross intersection point edge. The detection of this feature can effectively improve the accuracy and stability of component edge segmentation, thereby improving the robustness of subsequent anomaly detection algorithms and component positioning algorithms.

[0042] The present invention discloses a method for detecting the cross intersection points of the edges of photovoltaic components based on infrared images, including the following steps:

[0043] Step 1: Obtain infrared images of multiple scenarios through a drone, preprocess the infrared images, and eliminate high-frequency noise in the photovoltaic component area through mean filtering;

[0044] Step 2: Filter the preprocessed infrared images through a custom operator, detect the pixel areas that meet the requirements, and determine the cross intersection point candidate areas through Hough transform and morphological processing;

[0045] Step 3: Filter the pixels in the cross intersection point candidate areas through a custom operator to detect the cross intersection points in the photovoltaic component area.

[0046] Preferably, the specific content of Step 1 is: Obtain infrared images of multiple scenarios by shooting with a drone. The camera on the drone shoots along the direction of the bracket of the photovoltaic component. Set the line in the up-down direction in the infrared image as the longitude line, and the line in the left-right direction as the latitude line. The longitude line is long and the latitude line is short; control the height of the drone relative to the photovoltaic component and the equivalent focal length of the camera to ensure that the edge width of the photovoltaic component is 3 to 30 pixels.

[0047] The mean filtering is a prior art and will not be elaborated in this application.

[0048] Preferably, in Step 1, high-frequency noise in the photovoltaic component area is eliminated through mean filtering, and the mean filtering operator for mean filtering is 5*5.

[0049] Preferably, the multiple scenarios include mountain scenarios, water surface scenarios, roof scenarios, four-row plain scenarios, flat single-axis scenarios, and tilted single-axis scenarios.

[0050] Preferably, step 2 is specifically as follows:

[0051] Step 2-1: Filter through a custom operator, which is a 1*n matrix and includes three parts: the current pixel area, the component border area, and the component area. Filter each pixel through the custom operator to detect the pixel area of the warp feature of the photovoltaic component.

[0052] Step 2-2: Determine the cross intersection candidate area through Hough transform and morphological processing.

[0053] Preferably, the custom operator filtering in step 2-1 is as follows:

[0054] 2-1-1) Take the current pixel as the center, determine the window area range, and judge whether the current pixel value is the minimum value in the window area within this range. If not, directly jump out and execute the next pixel filtering, and indicate that the current pixel is not on the warp with a cross intersection. If satisfied, execute the next step;

[0055] 2-1-2) Compare the pixel values of each pixel in the component area with its adjacent pixels. If a large difference appears, immediately jump out and execute the next pixel filtering. If all the pixels in the component area meet the low difference requirement, it indicates that the current pixel meets the requirement, and the warp area image that meets the requirement is detected.

[0056] Preferably, the window area range is 20 to 30 pixels.

[0057] Preferably, the Hough transform and morphological operations in step 2-2 to determine the cross intersection candidate area are specifically as follows: Perform Hough transform on the warp area image to achieve linear fitting of the real warp area. The fitted line is redrawn and undergoes morphological dilation processing to obtain a continuous cross intersection candidate area.

[0058] The Hough transform and morphological processing are prior art and will not be elaborated in this application.

[0059] Preferably, step 3 includes the following steps:

[0060] 3-1) Use the cross intersection candidate area image as a mask image, and for each valid pixel in it, find the corresponding pixel position in the preprocessed infrared image;

[0061] 3-2) For this pixel position, determine a square area around it, and respectively determine four areas of up, down, left, and right on the four sides of the square area. These four areas are 1*n or n*1 pixel areas;

[0062] 3-3) Find the pixel with the minimum pixel value in each region;

[0063] 3-4) Calculate the position differences Δxt, Δxb, Δyl, and Δyr between the coordinates of the minimum pixels on the four sides and the center point of the square;

[0064] 3-5) If the differences between Δxt and Δxb and between Δyl and Δyr can both be guaranteed to be within the range of 10 pixels, then this pixel is considered the cross point.

[0065] Preferably, the square region in step 3-1 is a square.

[0066] Embodiment 1

[0067] The specific implementation steps of the method of the present invention are as follows:

[0068] Step 1:

[0069] Enlarge the cross region in the infrared image as shown in Figure 2 -c. The edge information of the component shows a lower brightness, and the battery cell region shows a higher brightness. Although there is a linear relationship between the brightness and temperature of the infrared image, during the actual inspection process, different weather, angles, and component models will all result in different image brightnesses. To improve the robustness of the algorithm and reduce the number of threshold settings, steps 2 and 3 will both process using the relative brightness between pixels. To improve the robustness of the subsequent steps, it is necessary to preprocess the input original image, and eliminate the high-frequency noise in other regions without affecting the relative brightness of the component edge region. The present invention uses the mean filtering algorithm to process the entire initial image in this link, maximizing the saving of processing time while ensuring the output quality.

[0070] Since the infrared images used in the present invention are all taken at a height of about 50m and the equivalent focal length of the camera is 58mm, the size of the component and the width of the edge can both be guaranteed within a certain scale range, that is, the edge width of the photovoltaic component is guaranteed to be within 3 to 30 pixels; the default setting of the mean filtering operator is 5*5, and this parameter can be adjusted if the camera and flight height are different.

[0071] As shown in Figure 2 shown, Figure 2 -b is a Figure 2 local enlarged view of Figure 2 -a, Figure 2 -c is a Figure 2 local enlarged view of

[0072] Step 2:

[0073] In the photovoltaic scene, in addition to the component foreground, the complex ground background will produce many interference features of suspected cross intersections, which increases the difficulty of accurate detection in step 3. In order to reduce the probability of false detection and improve the detection speed, this link will narrow the candidate area of ​​the cross intersection.

[0074] The implementation process is divided into two steps:

[0075] 2-1) Detect the pixel area of ​​the string warp feature, such as Figure 5 -b as shown;

[0076] 2-2) Determine the candidate cross intersection area through Hough transform and morphological processing, such as Figure 5 -c.

[0077] Since the present invention requires the camera to shoot along the direction of the bracket during the image acquisition process, for the convenience of explanation, the present invention defines the lines running up and down in the image as longitudes, and the lines running left and right as latitudes, with longitudes being longer and latitudes being shorter. According to the image display, the cross intersection must appear in the longitude or latitude area. Since the longitude is longer than the latitude and has stronger anti-interference ability, the present invention chooses to perform pixel-level segmentation on the longitude area.

[0078] Use a custom operator to filter the result of step 1. The operator is a 1*n matrix, which includes three areas: the current pixel area, the component border area, and the component area. Each pixel is filtered by the operator. The steps are as follows:

[0079] 2-1-1) Taking the current pixel as the center, determine the window area range, and determine whether the current pixel value is the minimum value in the window within the range. If it does not meet the requirement, directly jump out to execute the next pixel filtering, and indicate that the current pixel is not on the meridian with a cross intersection. If it meets the requirement, execute the next step;

[0080] 2-1-2) Compare the pixel values ​​of each pixel in the component area with its adjacent pixels. If a large difference occurs, immediately jump out and perform the next pixel filtering. If all pixels in the component area meet the low difference requirement, it means that the current pixel meets the requirement. After the entire image is filtered, the following is obtained Figure 5 -b's warp zone.

[0081] There is a lot of background interference in the meridian area image, and some meridian areas cannot be found in the previous step due to scale and resolution issues.

[0082] The present invention adopts the idea of ​​transformation statistics and performs Hough transform on the meridian area image, which not only realizes the straight line fitting of the real meridian area, but also eliminates the interference of the background area; the fitted straight line is redrawn and subjected to morphological expansion processing to obtain continuous cross intersection candidate areas, such as Figure 5as shown in -c.

[0083] Step 3:

[0084] The candidate regions generated in Step 2 narrow the scope of cross intersection detection and reduce the probability of false detection. In this step, for each pixel in the candidate region, cross intersection detection is directly performed on the preprocessed image in the Figure 4 way of -e.

[0085] As Figure 4 shown, with the cross intersection as the center, draw a square around it. The edges of the cross intersection and the sides of the square will show various intersection methods. Among them, except Figure 4 -b, the others will all have intersections with the sides. Since the present invention shoots along the bracket direction during image acquisition, the intersections of the cross intersection and the square can be shrunk within Figure 4 the three states of a, c, and d in.

[0086] The specific implementation steps are as follows:

[0087] 3-1) Use the candidate region image as a mask image. For each valid pixel in it, find the corresponding pixel position in the infrared image preprocessed in Step 1;

[0088] 3-2) For this pixel position, determine a square region around it, and respectively determine the upper, lower, left, and right four regions on the four sides of the square region. These four regions are pixel regions of 1*n or n*1;

[0089] 3-3) Find the pixel with the minimum pixel value in each region;

[0090] 3-4) Calculate the position differences Δxt, Δxb, Δyl, and Δyr between the minimum pixel coordinates on the four sides and the center point of the square;

[0091] 3-5) If the differences between Δxt and Δxb and the differences between Δyl and Δyr can both be guaranteed to be within the range of 10 pixels, then this pixel point is considered a cross intersection point. The final processing result is as Figure 5 shown in -d.

[0092] Figure 5 The result in -d needs to detect the cross in the 5-a image and requires the template in 5-c. Only the region in 5-a corresponding to the white region in 5-c participates in the cross detection, and the black region does not participate.

[0093] The principle of the present invention is as follows:

[0094] As Figures 1 to 5As shown, a method for detecting the cross intersection at the edge of a photovoltaic module based on an infrared image proposed by the present invention first preprocesses the input original image. While ensuring that the relative brightness of the edge region of the photovoltaic module remains unchanged, high-frequency noise in other regions is removed, improving the robustness of subsequent steps and reducing the number of threshold settings. Then, a custom operator is used for filtering, Hough transform, and morphological processing, narrowing the candidate region of the cross intersection, further reducing the probability of false detection, and improving the detection speed. Finally, the cross intersection is directly detected on the preprocessed image by filtering each pixel in the candidate region of the cross intersection using a custom operator. The present invention can achieve rapid detection of the cross intersections formed by the edges of adjacent modules in an infrared image, showing good generality and accuracy in various scenarios. The detection of the cross intersections provides assistance for subsequent links in photovoltaic inspection, effectively reducing the difficulty of component segmentation, component detection, and component positioning, etc.

[0095] The present invention proposes a method for detecting the cross intersection at the edge of a photovoltaic module based on an infrared image. This detection method relies on real-scene data of various types, combines the influences of various factors such as light, noise, and morphology, analyzes the commonalities and differences shown in the image by sorting out the real data of various scenarios, and extracts the idea of detecting the cross intersection. Since the metal frame at the edge of the photovoltaic module and the battery chip have different temperatures, in many scenarios, the infrared image can present relatively stable cross intersection edges. The detection of this feature can effectively improve the accuracy and stability of component edge segmentation, thereby improving the robustness of subsequent anomaly detection algorithms and component positioning algorithms, and reducing the difficulty of subsequent detection algorithms and positioning algorithms.

[0096] The present invention first processes the entire initial infrared image using the mean filtering algorithm. While ensuring that the relative brightness of the edge region of the photovoltaic module remains unchanged, high-frequency noise in other regions is removed, ensuring the output quality, maximizing the saving of processing time, and improving the robustness of subsequent steps.

[0097] The present invention uses a custom operator for filtering, as well as Hough transform and morphological processing, narrowing the candidate region of the cross intersection, further reducing the probability of false detection, and improving the detection speed.

[0098] Finally, the present invention directly detects the cross intersection on the preprocessed image by filtering each pixel in the candidate region of the cross intersection using a custom operator. By comparing the narrowed candidate region of the cross intersection with the preprocessed image, the accuracy is ensured while the efficiency is improved, effectively improving the robustness of subsequent anomaly detection algorithms and component positioning algorithms.

[0099] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those of ordinary skill in the art.

[0100] Many other changes and modifications can be made without departing from the concept and scope of the present invention. It should be understood that the present invention is not limited to specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A method for detecting the edge cross points of a photovoltaic module based on infrared images, characterized in that, It includes the following steps: Step 1: Obtain infrared images of multiple scenarios by using a drone, preprocess the infrared images, and remove high-frequency noise in the photovoltaic module area through mean filtering; Step 2: Filter the preprocessed infrared images through a custom operator, detect pixel regions that meet the requirements, and determine the cross intersection candidate regions through Hough transform and morphological processing; Step 3: Filter the pixels in the cross intersection candidate regions through a custom operator to detect the cross intersections in the photovoltaic module area; The said Step 3 includes the following steps: 3-1) Use the cross intersection candidate region image as a mask image, and for each valid pixel in it, find the corresponding pixel position in the infrared image preprocessed in Step 1; 3-2) For this pixel position, determine a square region around it, and respectively determine four regions of up, down, left, and right on the four sides of the square region. These four regions are pixel regions of 1*n or n*1; 3-3) Find the pixel with the minimum pixel value in each region; 3-4) Calculate the position differences Δxt, Δxb, Δyl, and Δyr between the coordinates of the minimum pixels on the four sides and the center point of the square; 3-5) If the differences between Δxt and Δxb and between Δyl and Δyr can both be guaranteed to be within the range of 10 pixels, then consider this pixel point as the cross intersection point; The said custom operator filtering is as follows: Take the current pixel as the center, determine the window region range, and judge whether the current pixel value is the minimum value in the window region within this range. If it does not meet the requirement, directly jump out and execute the filtering of the next pixel, and indicate that the current pixel is not on the meridian line with a cross intersection. If it meets the requirement, then execute the next step; Compare the pixel values of each pixel in the component region with its adjacent pixels. If a large difference appears, immediately jump out and execute the filtering of the next pixel. If all the pixels in the component region meet the requirement of low difference, it indicates that the current pixel point meets the requirement, and detect the meridian region image that meets the requirement; 2. The method for detecting the cross intersection points at the edges of a photovoltaic module based on an infrared image according to claim 1, wherein The said Step 1 is specifically: Obtain infrared images of multiple scenarios by using a drone to shoot. The camera on the drone shoots along the direction of the bracket of the photovoltaic module. Set the lines in the up and down directions in the infrared image as the meridians, and the lines in the left and right directions as the latitudes. The meridians are long and the latitudes are short; control the height of the drone relative to the photovoltaic module and the equivalent focal length of the camera to ensure that the edge width of the photovoltaic module is 3 to 30 pixels; 3. A method for detecting the edge cross points of a photovoltaic module based on an infrared image according to claim 2, wherein, In the said Step 1, high-frequency noise in the photovoltaic module area is removed through mean filtering, where a 5*5 mean filtering operator is used; 4. A method for detecting the edge cross points of a photovoltaic module based on an infrared image according to claim 2, characterized in that: The said multiple scenarios include mountain scenarios, water surface scenarios, roof scenarios, four-row plain scenarios, flat single-axis scenarios, and inclined single-axis scenarios; 5. A method for detecting the edge cross points of a photovoltaic module based on an infrared image according to claim 2, characterized in that, The said Step 2 is specifically: Step 2-1: Filter through a custom operator. The operator is a 1*n matrix, which includes three parts of regions: the current pixel region, the component border region, and the component region. Filter each pixel through the custom operator to detect the pixel point region of the photovoltaic module meridian characteristics; Step 2-2: Determine the cross intersection candidate regions through Hough transform and morphological processing; 6. The method for detecting the cross intersection points at the edges of a photovoltaic module based on an infrared image according to claim 1, wherein The said window region range is 20 to 30 pixels; 7. A method for detecting the edge cross points of a photovoltaic module based on an infrared image according to claim 5, characterized in that, In step 2-2, the specific method for determining the cross intersection candidate region by Hough transform and morphological operation is as follows: perform Hough transform on the longitude region image to achieve linear fitting of the real longitude region. The fitted line is redrawn and processed by morphological dilation, thus obtaining a continuous cross intersection candidate region.

8. A method for detecting the edge cross points of a photovoltaic module based on an infrared image according to claim 1, characterized in that: In step 3-1, the square region is a square.

Images