Photovoltaic module hot spot fault early warning method and system based on intelligent unmanned aerial vehicle patrol

Abstract

The invention belongs to the technical field of photovoltaic power station operation and maintenance, and relates to a photovoltaic module hot spot fault early warning method and system based on intelligent unmanned aerial vehicle patrol. Acquiring an ultraviolet image from the upper part of the photovoltaic module by using an unmanned aerial vehicle carrying an ultraviolet sensor; converting the ultraviolet image into a gray level image, for the gray level image, acquiring a gray level value of each pixel, and marking an area of which the gray level value is greater than or equal to a first threshold value as a highlight candidate area; calculating a gradient value of each highlight candidate area, and deleting the highlight candidate areas of which the gradient values are less than or equal to a second threshold value; judging whether each remaining highlight candidate region simultaneously meets the screening condition of the hot plate fault latency period or not; marking each highlight candidate area meeting the screening condition as a hot spot fault latent area; acquiring actual physical coordinates of each hot spot fault latent area; and generating an early warning signal of each hot spot fault latent area according to the actual physical coordinates, thereby realizing early warning of the hot spot fault of the photovoltaic module.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic power plant operation and maintenance technology, specifically relating to a method and system for early warning of hot spot faults in photovoltaic modules based on intelligent drone inspection. Background Technology

[0002] "Hot spots" on photovoltaic (PV) modules refer to a phenomenon where the temperature of a localized area of ​​the module is significantly higher than its surroundings, forming an abnormally high-temperature point. This phenomenon not only reduces power generation efficiency but may also lead to permanent damage to the module or even safety accidents. There are several reasons for hot spots on PV modules, one important reason being cracks in the glass cover of the solar cells. Cracks in the glass cover reduce the light transmittance of the glass, decreasing the amount of sunlight reaching the solar cells. This causes the solar cells to become energy-consuming loads, consuming the energy generated by other normal solar cells, which in turn leads to a localized increase in temperature within the PV module, resulting in hot spots.

[0003] Currently, the common method for detecting hot spot faults in photovoltaic (PV) modules is to use infrared thermal imagers to acquire temperature distribution images of the PV module surface. Analysis of abnormal temperature areas in the infrared image is then used to determine whether a hot spot fault has occurred and to pinpoint its location. However, when the crack in the glass cover is only a shallow crack, it is insufficient to significantly increase the local resistance of the solar cell, and the heat generation is below the hot spot threshold. Therefore, no obvious hot spot will appear at the location corresponding to the shallow crack in the infrared image, making it easily overlooked. However, shallow cracks in the glass cover can gradually develop into deep cracks or even through cracks under the influence of external environmental factors (temperature changes, external forces, etc.). If these cracks are not addressed in their early stages, hot spot faults will appear in the later stages, and even if the hot spot is detected by an infrared thermal imager, there will be a significant lag. Therefore, it is necessary to detect shallow cracks in the glass cover of PV modules to identify potential hot spot faults and achieve early warning of hot spot faults in PV modules. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: Firstly, a method for early warning of hot spot faults in photovoltaic modules based on intelligent drone field inspection is proposed, including the following steps: A drone equipped with an ultraviolet sensor collects ultraviolet images from above the photovoltaic modules; the ultraviolet images are converted into grayscale images, and the following steps are performed on the grayscale images: the grayscale value of each pixel is obtained, and regions with grayscale values ​​≥ a first threshold are marked as bright spot candidate regions; the gradient value of each bright spot candidate region is calculated, and bright spot candidate regions with gradient values ​​≤ a second threshold are deleted; it is determined whether each remaining bright spot candidate region simultaneously satisfies conditions one to three; wherein, condition one: the aspect ratio of the bright spot candidate region is ≥ a third threshold, condition two: the circularity of the bright spot candidate region is ≤ a fourth threshold, and condition three: the number of broken pixels in the bright spot candidate region is ≤ the total number of pixels in the bright spot candidate region × a fifth threshold; each bright spot candidate region that simultaneously satisfies conditions one to three is marked as a hot spot fault latent region; the actual physical coordinates of each hot spot fault latent region are obtained; and an early warning signal for each hot spot fault latent region is generated based on the actual physical coordinates.

[0005] Secondly, a photovoltaic module hot spot fault early warning system based on intelligent UAV field inspection is proposed, comprising: an image acquisition module for acquiring ultraviolet images from above the photovoltaic module; an image processing module for converting the ultraviolet images into grayscale images; a first data analysis and processing module for obtaining the grayscale value of each pixel and marking regions with grayscale values ​​≥ a first threshold as bright spot candidate regions; a second data analysis and processing module for calculating the gradient value of each bright spot candidate region and deleting bright spot candidate regions with gradient values ​​≤ a second threshold; a third data analysis and processing module for determining whether each remaining bright spot candidate region simultaneously satisfies conditions one to three, and marking each bright spot candidate region that simultaneously satisfies conditions one to three as a hot spot fault latent region; wherein, condition one: the aspect ratio of the bright spot candidate region ≥ the third threshold, condition two: the circularity of the bright spot candidate region ≤ the fourth threshold, and condition three: the number of broken pixels in the bright spot candidate region ≤ the total number of pixels in the bright spot candidate region × the fifth threshold; a physical coordinate acquisition module for obtaining the actual physical coordinates of each hot spot fault latent region; and an early warning signal generation module for generating an early warning signal for each hot spot fault latent region based on the actual physical coordinates.

[0006] Compared with existing technologies, this invention has the following advantages and beneficial effects: This method avoids the deficiency of photovoltaic hot panel monitoring technology based on infrared thermal imaging in being unable to provide early warning of potential hot spot faults caused by shallow cracks. It achieves early warning of hot spot faults in photovoltaic modules by analyzing and processing ultraviolet images of the photovoltaic module surface. Specifically, by acquiring ultraviolet images of the photovoltaic module and combining them with the characteristics of shallow cracks on the photovoltaic module's glass cover, the ultraviolet images are analyzed and processed to mark the potential hot spot fault areas representing shallow cracks in the ultraviolet images. The actual physical location of the shallow cracks on the photovoltaic module's glass cover is also obtained, enabling the investigation of shallow cracks on the photovoltaic module's glass cover. An early warning signal is generated based on the actual physical location of the shallow cracks, promptly reminding the maintenance team to address the shallow cracks and prevent them from developing into deep cracks or even through cracks, thus avoiding hot spot faults in the photovoltaic module. Attached Figure Description

[0007] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of a photovoltaic module hot spot fault early warning method based on intelligent drone field inspection provided in Embodiment 1 of the present invention.

[0008] Figure 2 This is a schematic diagram of the overall method flow for distortion correction and geometric correction of grayscale images provided in Embodiment 1 of the present invention.

[0009] Figure 3 This is a schematic flowchart of a method for correcting radial and tangential distortion of a grayscale image according to Embodiment 1 of the present invention.

[0010] Figure 4 A schematic flowchart of the method for obtaining three-dimensional coordinates during ultraviolet sensor imaging provided in Embodiment 1 of the present invention.

[0011] Figure 5 A schematic flowchart of the method for obtaining the actual physical coordinates of the hot spot fault latency zone provided in Embodiment 1 of the present invention. Detailed Implementation

[0012] This explanation is intended to clarify the invention and is not intended to limit it. The embodiments described below are some, but not all, of the embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0013] In the following description, numerous specific details are set forth to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other embodiments, well-known structures, materials, or methods are not specifically described to avoid obscuring the invention. Unless otherwise specified, the materials, instruments, and reagents used in the following embodiments are commercially available. Unless otherwise specified, the techniques used in the embodiments are conventional methods well known to those skilled in the art.

[0014] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0015] Example 1: A method for early warning of hot spot faults in photovoltaic modules based on intelligent UAV field inspection is proposed. During the UAV field inspection, ultraviolet images above the photovoltaic modules are collected. Combined with the characteristics of shallow cracks on the glass cover of the photovoltaic modules, the ultraviolet images are analyzed and processed to mark the potential hot spot fault areas representing shallow cracks in the ultraviolet images. The actual physical location of the shallow cracks on the glass cover of the photovoltaic modules is also obtained, enabling the inspection of shallow cracks on the glass cover of the photovoltaic modules. An early warning signal is generated based on the actual physical location of the shallow cracks to promptly remind the operation and maintenance personnel to deal with the shallow cracks and prevent them from developing into deep cracks or even through cracks, which would lead to hot spot faults in the photovoltaic modules.

[0016] Based on the above technical approach, this method specifically includes: Figure 1 The following implementation steps are shown: Step 1: Use a drone equipped with an ultraviolet sensor to collect ultraviolet images from above the photovoltaic module.

[0017] Explanation of the necessity of acquiring ultraviolet images: 1. The core structure of a photovoltaic module includes a glass cover, encapsulation materials, solar cells, and a backsheet. Among these, the glass cover is the main carrier for reflecting ultraviolet light, and its reflectivity for ultraviolet light is significantly higher than that for visible light.

[0018] 2. When the glass cover is intact, the reflection of ultraviolet light mainly comes from specular reflection on the glass surface, with concentrated reflection direction and uniform energy distribution. However, the appearance of cracks disrupts the continuity of the glass surface, forming "interface defects." This causes the glass cover to change its ultraviolet light reflection from specular reflection to scattered reflection. Furthermore, multiple interfaces exist between air and glass within the crack (the crack gaps are filled with air). Ultraviolet light undergoes multiple reflections and refractions at these interfaces, ultimately increasing the reflected energy escaping from the glass surface. This results in a stronger ultraviolet reflection signal that can be detected at non-specular angles. Therefore, when shallow cracks on the glass cover cannot be detected by hot spots in infrared images, a drone equipped with an ultraviolet sensor can be used to collect ultraviolet light reflection signals above the glass cover. By analyzing and processing the ultraviolet images, the shallow cracks on the glass cover and their locations can be detected.

[0019] Step 2: Convert the ultraviolet image to a grayscale image.

[0020] Based on the above description of the overall technical approach, after acquiring the ultraviolet image above the photovoltaic module, it is necessary to analyze and process the ultraviolet image in conjunction with the shallow crack features on the glass cover to identify the shallow cracks on the glass cover. The identification method compares and analyzes the grayscale values ​​of the pixels with the feature thresholds of the shallow cracks; therefore, the ultraviolet image needs to be converted into a grayscale image before identification.

[0021] It should be noted that after converting the ultraviolet image to a grayscale image, the following steps are also included: Step 2.1: Perform filtering and histogram equalization on the grayscale image.

[0022] Grayscale images may contain sensor noise (such as dark current and random noise) and environmental interference (such as ultraviolet reflection from the sky and reflection from adjacent components). Therefore, filtering algorithms (such as Gaussian filtering and median filtering) are needed to remove noise from grayscale images. To further enhance the brightness difference between shallow cracks and the background, histogram equalization can be applied to the filtered grayscale image, making the shallow cracks appear as clearer "high-brightness linear areas" and creating a more obvious distinction from the surrounding low-brightness background.

[0023] Step 2.2: Perform distortion correction and geometric correction on the grayscale image processed in Step 2.1.

[0024] When taking photos with a drone, the ultraviolet images may be distorted due to the tilt angle (such as a downward angle of 30°~60°) (e.g., objects appear larger when closer and smaller when farther away, or edges are stretched). Therefore, it is necessary to combine the actual size of the photovoltaic module (e.g., a standard module of 1640mm×992mm) and GPS positioning data to correct the converted grayscale image into an orthophoto (frontal view). This ensures that the position of the shallow crack in the grayscale image corresponds one-to-one with its actual position on the glass cover, thereby avoiding misjudgment of the actual position of the shallow crack on the glass cover due to perspective distortion.

[0025] Step 2.2 includes Figure 2 The following steps are shown: Step 2.2.1: Use the distortion coefficient of the ultraviolet sensor to perform radial distortion correction and tangential distortion correction on the grayscale image.

[0026] Before proceeding, it is necessary to obtain the basic parameters of the photovoltaic module, the calibration parameters of the ultraviolet sensor, and the installation parameters of the ultraviolet sensor. The basic parameters of the photovoltaic module include: the standard physical dimensions of the photovoltaic module (e.g., 1640mm × 992mm), the installation tilt angle, the number of rows and columns, the spacing between the cells, the actual coordinates of the four corner points, and the pixel coordinates of the lens optical center in the ultraviolet image. The calibration parameters of the ultraviolet sensor include: the focal length, principal point coordinates, and distortion coefficient of the ultraviolet sensor. The installation parameters of the ultraviolet sensor include: the three-dimensional coordinate offset of the ultraviolet sensor relative to the center of gravity of the UAV (offset in the x, y, and z directions) and the angle between the ultraviolet sensor and the three-dimensional coordinate system of the UAV.

[0027] It should be noted that the purpose of collecting the basic parameters of the photovoltaic module is to provide physical constraints for the subsequent calculation of the actual physical location of the shallow crack on the glass cover plate; the purpose of collecting the calibration intrinsic parameters of the ultraviolet sensor is to eliminate the influence of lens optical distortion on the image; and the purpose of collecting the installation extrinsic parameters of the ultraviolet sensor is to ensure that the shooting direction of the ultraviolet sensor is consistent with the coordinate system of the UAV.

[0028] Based on the fundamental parameters of the photovoltaic module, the calibration intrinsic parameters of the ultraviolet sensor, and the installation extrinsic parameters of the ultraviolet sensor, radial and tangential distortion corrections are performed on the grayscale image using the distortion coefficients of the ultraviolet sensor. This eliminates image distortion caused by lens optical characteristics (such as edge stretching, pincushion / barrel distortion), resulting in a "lens-distortion-free original image." Essentially, this involves establishing a mathematical mapping relationship between distorted pixel coordinates and ideal distortion-free coordinates using the distortion coefficients in the camera's intrinsic parameters, thus "restoring" the distorted image to the real scene. The specific method involves performing the following steps on each pixel in the grayscale image: Figure 3 Steps A1 to A4 are shown.

[0029] Step A1: Convert the distorted coordinates of the pixels into normalized image coordinates.

[0030] The formula is: x =( u - c x ) / f x , y =( v - c y ) / f y .in,( x , y ) represents the normalized pixel coordinates, u , v ) represents the coordinates of the distorted pixels, ( c x , c y () represents the pixel coordinates of the optical center of the ultraviolet sensor lens in the ultraviolet image. f x For ultraviolet sensors in x Focal length of direction, f y For ultraviolet sensors in y Focal length in direction.

[0031] Step A2: Input the normalized pixel coordinates, the radial distortion coefficient of the ultraviolet sensor, and the square of the normalized distance from the pixel to the lens optical center into the radial correction formula, and output the pixel coordinates after radial distortion correction.

[0032] The radial correction formula is: .in,( x radial , y radial () represents the pixel coordinates after radial distortion correction. r 2 That is, the square of the normalized distance from the pixel to the lens optical center. r 2 = x 2 + y 2 , k 1. k 2 and k 3 represents the radial distortion coefficient of the ultraviolet sensor.

[0033] Step A3: Input the tangential distortion correction formula with the distorted pixel coordinates, normalized pixel coordinates, tangential distortion coefficient of the ultraviolet sensor, and normalized distance from the pixel to the lens optical center, and output the tangential distortion correction pixel coordinates.

[0034] The tangential correction formula is: .in,( x tangential , y tangential () represents the pixel coordinates after tangential distortion correction. p 1 and p 2 are both tangential distortion coefficients.

[0035] Step A4: Convert the pixel coordinates after radial distortion correction and tangential distortion correction back to pixel coordinates.

[0036] The conversion formula is: .in,( u undistorted , v undistorted () represents the coordinates of the undistorted pixel.

[0037] Step 2.2.2: Based on the GPS data recorded by the UAV during shooting and the three-dimensional coordinate offset of the ultraviolet sensor from the center of gravity of the UAV, obtain the three-dimensional coordinates of the ultraviolet sensor during shooting.

[0038] The GPS data recorded by the drone reflects the three-dimensional coordinates of the drone's center of gravity. x drone , y drone , z drone The position of the ultraviolet sensor needs to be based on the relative three-dimensional offset between the two (Δ). x ,△ y ,△ z Since the UAV has attitude angles (heading angle, pitch angle, roll angle) during flight, the offset needs to be transformed from the UAV body coordinate system to the ground coordinate system through attitude transformation (coordinate rotation) to finally obtain the absolute three-dimensional coordinates of the sensor. It should be noted that: (1) The coordinates output by the UAV GPS are expressed in latitude and longitude (Lat, Lon) and altitude (Alt), which need to be converted into three-dimensional rectangular coordinates (x, y, z, unit: meters) for calculation; (2) The origin of the three-dimensional coordinate system of the ultraviolet sensor is the center of gravity of the UAV, the x-axis is forward along the body, the y-axis is to the right, and the z-axis is downward (right-hand coordinate system), which is used to define the offset of the sensor relative to the center of gravity.

[0039] The specific implementation steps for obtaining the three-dimensional coordinates during ultraviolet sensor imaging are as follows: Figure 4 As shown, it includes: Step B1: Convert the GPS data (latitude, longitude, and altitude) recorded by the drone into a geodetic rectangular coordinate system.

[0040] The conversion formula is: Where N is the radius of curvature of the zonal loop. ,a The length of the Earth's semi-major axis. a =6378137m, e 2 Let the square of the Earth's ellipsoid eccentricity be the eccentricity. e 2 =0.00669437999014, ( Lat d , Lon d , Alt d () represents the geodetic coordinates taken by the drone.

[0041] It should be noted that, Lat d and Lon d， It needs to be converted to radians.

[0042] Step B2: Obtain the three-dimensional offset of the ultraviolet sensor relative to the center of gravity of the UAV.

[0043] The offset of the ultraviolet sensor relative to the center of gravity of the drone needs to be (△) x ,△ y ,△ z The measurement is performed in the fuselage coordinate system. The measurement method is as follows: The distance between the installation position of the ultraviolet sensor and the center of gravity of the drone is obtained from the drone's mechanical design drawings (e.g., along the fuselage forward Δ...). x= 0.5m, to the right △ y= 0.2m, downward △ z= 0.1m).

[0044] Step B3: Convert the three-dimensional offset to the geodetic coordinate system using the attitude angle of the UAV.

[0045] Attitude angles (heading angles) during drone flight Pitch angle Roll angle This will cause a rotational relationship between the fuselage coordinate system and the geodetic coordinate system, and the offset in the fuselage coordinate system needs to be converted to the geodetic coordinate system using a rotation matrix.

[0046] (1) Convert the attitude angles of the UAV during flight into the corresponding rotation matrix.

[0047] The rotation sequence is usually zyx (yaw angle - pitch angle - roll angle), and the corresponding rotation matrix R is: Among them, the roll angle rotation matrix Pitch angle rotation matrix yaw angle rotation matrix .

[0048] (2) Based on the rotation matrix, convert the offset of the ultraviolet sensor relative to the UAV's center of gravity to the offset in the geodetic coordinate system. The conversion formula is: .in, This represents the offset of the ultraviolet sensor relative to the center of gravity of the drone. This represents the three-dimensional offset of the ultraviolet sensor relative to the center of gravity of the UAV in the geodetic coordinate system.

[0049] Step B4: Superimpose the coordinates of the UAV's center of gravity with the three-dimensional offset of the ultraviolet sensor relative to the UAV's center of gravity in the geodetic coordinate system to obtain the three-dimensional coordinates of the ultraviolet sensor during shooting.

[0050] Step 2.2.3: Establish a mapping relationship model between the pixel coordinate system of the grayscale image and the three-dimensional coordinate system of the photovoltaic module.

[0051] The expression for the mapping relationship model is: ;in,( u , v () represents the actual pixel coordinates of a point on the surface of the photovoltaic module in the ultraviolet image. X , Y , Z () represents the actual three-dimensional coordinates of a point on the surface of a photovoltaic module. X s , Y s , Z s () represents the three-dimensional coordinates when the ultraviolet sensor takes a picture. f Indicates the focal length of the ultraviolet sensor, ( u 0, v 0) represents the coordinates of the principal point of the ultraviolet sensor.

[0052] Step 2.2.4: Input the actual three-dimensional coordinates of each point on the surface of the photovoltaic module, the three-dimensional coordinates when the ultraviolet sensor takes the picture, the focal length of the ultraviolet sensor, and the principal point coordinates of the ultraviolet sensor into the mapping relationship model, and output the actual pixel coordinates of each point on the surface of the photovoltaic module in the grayscale image.

[0053] Step 2.2.5: Perform coordinate correction on each pixel in the grayscale image using the actual pixel coordinates.

[0054] The coordinate correction described in this step involves modifying the current coordinates of each pixel in the grayscale image using the actual pixel coordinates of each point on the surface of the photovoltaic module in the grayscale image.

[0055] Step 2.2.6: Calculate the ratio between the actual physical size of the photovoltaic module frame and the pixel size of the photovoltaic module frame in the grayscale image to obtain the size conversion factor.

[0056] For example, for a complete photovoltaic (PV) module captured in a grayscale image, the pixel dimensions (e.g., width W_pixel, height H_pixel) of the PV module's edges in the grayscale image are determined by detecting the module's edges (e.g., rectangular outline). The size conversion factor then includes a width conversion factor W1 and a length conversion factor W2. W1 = actual width of the PV module's frame / W_pixel, W2 = actual height of the PV module's frame / H_pixel.

[0057] It should be further explained that the method for detecting the edges of photovoltaic modules in grayscale images is as follows: the Canny edge detection algorithm and the Hough line detection algorithm are used to extract the bounding box of the photovoltaic modules from the grayscale image.

[0058] 1. The Canny edge detection algorithm is a gradient-based edge detection algorithm that detects edges in an image through a series of steps. The core steps of the Canny edge retrieval algorithm are: (1) Apply Gaussian filtering to the image to be detected. The purpose is to reduce noise in the image and avoid noise interference with edge detection. The method is to use a Gaussian filter to perform a convolution operation on the image (the kernel size and standard deviation of the Gaussian filter can be adjusted as needed).

[0059] (2) Calculate the gradient magnitude and direction. The purpose is to detect the gradient information of each pixel in the image and determine the direction and intensity of the edges. The method is to use the Sobel operator to calculate the gradient of the image to be detected in the horizontal direction. G x and the gradient in the vertical direction G y Then calculate the gradient magnitude. M and gradient direction θ The gradient magnitude represents the intensity of the edge, and the gradient direction represents the direction of the edge. , .

[0060] (3) Non-maximum consistency. The goal is to refine the edges and ensure that the edges are only one pixel wide. The method is to check the neighboring pixels in the gradient direction for each pixel. If the gradient magnitude of the current pixel is not a local maximum, then set its gradient magnitude to 0. This includes: (1) quantizing the gradient direction into 8 directions (0°, 22.5°, 45°, 67.5°, 90°, 112.5°, 135°, 157.5°); (2) for each pixel, check the two neighboring pixels in the gradient direction; (3) if the gradient magnitude of the current pixel is not the maximum of the two neighboring pixels, then set its gradient magnitude to 0.

[0061] (4) Dual threshold detection. The purpose is to distinguish between strong and weak edges by using high and low thresholds to avoid misjudgment. The method is (1) setting two thresholds: a high threshold and a low threshold. Th and low threshold Tl (1) A high threshold is used to detect strong edges, and a low threshold is used to detect weak edges; (2) Traverse the gradient magnitude image after non-maximum suppression - if the gradient magnitude is greater than the high threshold Th If the gradient magnitude is less than the low threshold, then the pixel is considered a strong edge and is marked as an edge pixel; Tl If the gradient magnitude is below a low threshold, then the pixel is considered not an edge and is marked as a non-edge pixel; Tl and high threshold Th If a pixel is between two edges, its eight neighboring pixels are examined. If there are strong edge pixels in the neighborhood, the pixel is also considered an edge pixel; otherwise, the pixel is not considered an edge pixel.

[0062] (5) Edge connection. The purpose is to connect broken edges to form continuous edges. The method is to traverse the edge image and connect weak edge pixels with strong edge pixels to form complete edges.

[0063] (6) Output edge image. The final generated edge image is represented by white pixels at the edges and black pixels at the non-edges.

[0064] 2. The Hough line detection algorithm is used to extract lines from edge images. It maps each edge point in the image to a curve in the parameter space, and uses an accumulator to count the intersections of these curves, thus detecting the line. The core steps of the Hough line detection algorithm are: (1) Define the size of the accumulator matrix and initialize it to a zero matrix. The accumulator matrix is ​​used to count the number of intersections of each point in the parameter space. If the number of intersections of a point exceeds a threshold, the parameter corresponding to that point is considered to represent a straight line.

[0065] (2) Traverse each edge point ( p , q For each possible Value, calculate the corresponding The value will be the corresponding value in the accumulator matrix. , The position is incremented by 1. In the Hough transform, each edge point ( p , q Each of these will generate a curve in the parameter space. The equation of this curve is: .in, Let be a variable, with values ​​ranging from 0° to 180°. For each edge point, all possible values ​​need to be calculated. corresponding Value. Additionally, in the parameter space, each ( , This corresponds to a cell in the accumulator matrix. By traversing all possible... The value can be used to count each ( , The number of intersections corresponding to ).

[0066] (3) Traverse the accumulator matrix and find the points whose values ​​exceed the threshold. These points correspond to ( , The parameter represents a straight line in the edge image.

[0067] (4) For each detected ( , ) parameters, calculate the intersection point of the line and the image boundary: .

[0068] Based on the Canny edge detection algorithm and the Hough line detection algorithm mentioned above, the method for extracting the outline of photovoltaic modules from grayscale images is as follows: First, the gradient of the grayscale image is calculated (e.g., by using the Sobel operator to detect grayscale changes in the horizontal and vertical directions). Since obvious gradient peaks can be formed at the dark lines of the border due to abrupt changes in grayscale values ​​(from a high grayscale surface to a low grayscale border), the edges of the photovoltaic module can be preliminarily identified by calculating the gradients of the grayscale image in the horizontal and vertical directions.

[0069] Then, by using a dual threshold method (high threshold to filter strong edges, low threshold to connect weak edges), continuous dark border lines are retained, while discrete noise edges (such as short edges of component surface scratches) are eliminated.

[0070] Finally, straight line segments in the edge are extracted using the Hough line detection algorithm. This includes: (1) setting a threshold for the length of the straight line (e.g., at least 80% of the length of the short side of the component to avoid misjudging it as a small scratch) and angle constraints (the component border is mostly a straight line that is horizontal / vertical or consistent with the installation tilt angle, and the angle deviation is usually <5°); (2) selecting 4 straight line segments that meet the rectangular characteristics (two parallel lines and adjacent lines perpendicular) as the four dark lines of the component border.

[0071] Step 2.2.7: Perform size correction on each pixel in the UV image according to the size conversion factor.

[0072] The correction described in this step involves scaling the width of each pixel using a size conversion factor in the width direction and scaling the height of each pixel using a size conversion factor in the height direction.

[0073] Step 3: Perform steps 4 through 8 on the grayscale image.

[0074] Before detailing the specific implementation methods of steps 4 to 8, a morphological analysis of the shallow cracks on the glass cover plate is necessary to extract their typical characteristics. Specifically: The morphology of shallow cracks in glass cover plates in ultraviolet images is closely related to the physical characteristics of the cracks (such as direction, width, and edge condition) and the scattering patterns of ultraviolet light, as follows: 1. Core Feature – Bright Linear or Striped Areas. Ultraviolet (UV) sensors are sensitive to reflected / scattered UV light. Shallow cracks, due to enhanced scattering of UV light (higher reflectivity than intact glass surfaces), will appear as brighter linear or striped features in UV images than surrounding areas; this is the most prominent identification characteristic. If the crack is a straight, shallow crack, it will appear as a continuous bright line in the UV image, with the line width corresponding to the actual crack width (but slightly wider than the actual size due to scattering, typically 1-3 pixels wide). If the crack has branches or bends, the bright area will extend along the crack's direction, forming forked or zigzag bright bands.

[0075] 2. Edge features - blurry and irregular. The edges of shallow cracks are usually uneven (with tiny jagged edges or debris), which makes the direction of ultraviolet light scattering more dispersed. Therefore, in the ultraviolet image: (1) the edges of the bright areas of shallow cracks will not show clear straight lines, but will show blurry "rough edges" or gradual transitions (the brightness gradually decreases from the center of the bright area to the surrounding normal area); (2) if dust or water vapor is embedded in the shallow crack, it will further enhance local scattering, and intermittent "spot-like bright spots" will appear in the bright lines, making the overall shape more irregular.

[0076] 3. The brightness difference between shallow cracks and the surrounding area is significant. On an intact glass cover, ultraviolet light is primarily reflected specularly, with a concentrated reflection direction. When the ultraviolet sensor collects light at a non-mirror angle, the surrounding area has low brightness (or a uniformly dark background). However, the scattering of light from shallow cracks increases the energy of ultraviolet light reflected towards the sensor, resulting in a significant brightness contrast between the cracked area and the surrounding background (typically a difference of 20% to 50%). Even extremely fine shallow cracks can be identified through this brightness difference.

[0077] 4. Shallow cracks present discontinuous or incomplete bright traces. (1) For very shallow and narrow shallow cracks (such as only micron-level damage on the surface), due to the weak scattering energy, they will appear as discontinuous bright spots or short lines in the image; (2) If the shallow crack is partially covered by dust, the ultraviolet scattering of the covered area will be weakened, which may cause the bright lines to "break" or "local darken", forming a discontinuous shape.

[0078] 5. Non-crack interference characteristics. (1) Compared with scratches on the glass surface, the bright lines of shallow cracks are more continuous and the edge blur is higher. Scratches are shallower and smoother, and the scattering is weaker. The brightness is usually lower than that of cracks and the edges are relatively clear. (2) Stains (such as bird droppings and dust accumulation) absorb or diffusely reflect ultraviolet light and usually present dark spots or low brightness areas, which is the opposite of the bright characteristics of shallow cracks. (3) Due to the series and parallel structure of the battery cells, they may present weak stripes, but the gray value fluctuation is small (±10) and there is no obvious bright area.

[0079] Based on the morphological analysis of the shallow cracks described above, areas in the grayscale image that meet the criteria of "highlight, blurred edges, linear / striped" can be identified as the locations of shallow cracks, while excluding non-crack interference (such as scratches, stains, junction box reflections, etc.). Refer to step 4 below.

[0080] Step 4: Obtain the grayscale value of each pixel, and mark the regions with grayscale values ​​greater than or equal to the first threshold as highlight candidate regions.

[0081] Because shallow cracks exhibit enhanced ultraviolet reflection, they appear as areas with higher grayscale values ​​than the surrounding background in the image (a core feature). Therefore, step 4 involves identifying areas in the grayscale image with grayscale values ​​≥ the lower limit of the grayscale threshold based on this core feature of the shallow cracks. T low The area is marked as "highlight candidate area".

[0082] Prior to this, it is necessary to conduct sample training (collecting ultraviolet images of photovoltaic modules containing shallow cracks), statistically analyze the grayscale value distribution in the crack area (usually between 180 and 255, which needs to be adjusted according to the parameters of the ultraviolet sensor), and set a lower limit for the grayscale threshold. T low (e.g., 180). By training samples and statistically analyzing the grayscale distribution of shallow crack regions in photovoltaic modules, the goal is to quantify the "highlight" characteristics of shallow cracks at the data level, providing an objective grayscale threshold or distribution model for subsequent crack detection. The specific implementation can be divided into the following five key steps: 1. Construct a sample dataset.

[0083] The quality of the sample directly determines the effectiveness of grayscale distribution statistics. It is necessary to cover the characteristics of shallow cracks in different scenarios to ensure the diversity and representativeness of the dataset.

[0084] (1) Sample collection conditions.

[0085] Use a drone identical to the one used in actual testing, equipped with an ultraviolet light sensor (such as an ultraviolet band CCD camera), and fix the focal length and exposure parameters (to avoid grayscale deviation due to parameter fluctuations).

[0086] The focus is on shallow cracks (depth < 10% of the thickness of the glass on the component surface, length 5mm~50mm, width < 0.5mm), and includes different forms (straight, bifurcated, and network-like shallow cracks).

[0087] Simultaneously acquire images containing non-crack interference, such as surface scratches (deep, sharp edges), stains (uneven grayscale blocks), junction box reflections (locally bright areas), and dust-covered areas (overall low grayscale), for subsequent differentiation.

[0088] Shoot under different lighting conditions (sunny noon, cloudy, early morning / evening) and different angles (30°~90° angle between the drone and the component surface) to ensure that the sample covers the differences in light reflection.

[0089] (2) Sample size.

[0090] The number of valid shallow crack samples should be ≥500 (each image can contain 1-3 cracks), with a balanced proportion of samples of different shapes and under different lighting conditions (e.g., straight cracks account for 40%, bifurcation cracks account for 30%, and mesh cracks account for 30%). The number of interfering samples should be ≥300 to ensure coverage of interfering features.

[0091] 2. Standardize the grayscale benchmark for image samples.

[0092] Raw ultraviolet images may contain noise and grayscale shifts, requiring preprocessing to eliminate irrelevant interference and unify grayscale scales. This includes: (1) Image denoising. Ultraviolet images are susceptible to sensor noise (such as salt and pepper noise). Gaussian filtering (smoothing high-frequency noise) or median filtering (preserving edges while removing impulse noise) is used. The parameters are adjusted according to the noise intensity (such as Gaussian kernel size 3×3~5×5).

[0093] (2) Grayscale normalization. Convert the color ultraviolet image (if the sensor outputs RGB format) to a grayscale image (single channel). Unify the grayscale scale under different exposure conditions by linear stretching (e.g., mapping pixel values ​​to the range of 0~255). The formula is: Grayscale value = ( II min ) / ( I max - I min )×255. Among them, I These are the original pixel values. I min The minimum global pixel value of the image. I max This is the global maximum pixel value of the image.

[0094] 3. Mark the cracked areas.

[0095] Use professional image annotation tools (such as LabelMe, VGGImageAnnotator) to use polygon annotation or pixel-level mask annotation. Note that for linear shallow cracks, polygons should be drawn along the crack edge to cover all pixels in the crack area (including blurred edges); when annotating, it is necessary to strictly distinguish between cracks and interference areas (e.g., label cracks as "crack" and interference areas as "noise").

[0096] 4. Extract the grayscale values ​​of the shallow crack area.

[0097] Pixel-level grayscale values ​​are extracted from the marked shallow crack areas to form the original data sample. This includes: (1) Extracting the Region of Interest (ROI). Based on the labeled mask file, extract the ROI of the shallow crack area from the preprocessed grayscale image, obtain the grayscale values ​​of all pixels within this area, and set it as a set. G crack ={ g 1, g 2,..., g n}, n This represents the total number of pixels in the shallow crack area.

[0098] (2) Extract the grayscale values ​​of the interference areas. Simultaneously extract the grayscale value set of the labeled non-crack interference areas (scratches, stains, etc.). G noise This is used for subsequent comparative analysis.

[0099] (3) Extracting grayscale values ​​from normal areas. Select normal component surface areas without cracks or interference in the image and extract the set of grayscale values. G normal , as a benchmark reference.

[0100] 5. Perform statistical analysis on the distribution of gray values.

[0101] Statistical analysis G crack The distribution characteristics were analyzed to clarify the grayscale range of shallow cracks and their differences from other areas. This included: (1) Calculate the mean gray value of the crack region ( ), median ( m crack ), standard deviation ( ), minimum value ( g min,crack ) and maximum value (g) max,crackThis describes the overall brightness level and dispersion. For example, if the average gray level of a normal area is 80 (range 0~255), the average gray level of a shallow crack area may reach 150~200 due to enhanced ultraviolet reflection, and the standard deviation is smaller (the brightness distribution is more uniform).

[0102] (2) Draw a histogram. Plot the grayscale value (0~255) on the horizontal axis and the number of pixels on the vertical axis. G crack , G noise and G normal The grayscale histogram visually illustrates the distribution differences among the three.

[0103] Shallow crack histograms typically exhibit a single-peak distribution, with peak values ​​concentrated in the high grayscale range (e.g., 160~220) and relatively narrow peak values ​​(due to more uniform reflectance).

[0104] The peak values ​​of the histogram in the normal region are concentrated in the low to medium grayscale range (e.g., 60~100).

[0105] The scratched area may show a bimodal distribution due to edge reflection (brighter edges and darker interior), while the histogram of the stained area is scattered (with no obvious peak).

[0106] (3) Fit the probability density function. Use kernel density estimation (KDE) to... G crack By fitting the data, a continuous grayscale distribution curve is obtained, and the confidence interval of the crack grayscale (such as the grayscale range with a 95% probability) is determined.

[0107] (4) Determine the threshold range. This is done through comparison. G crack and G normal The distribution of the grayscale values ​​was used to determine the lower limit of the grayscale threshold for shallow cracks. T low (Maximum grayscale value in the normal area) and upper limit of grayscale threshold T high（ For example, setting the grayscale threshold to 240 is to avoid confusion with the strong reflection from the junction box. For instance, if the grayscale of the normal area is ≤100, and the grayscale distribution of the shallow crack area is between 180 and 220, then a preliminary setting can be made. T low =180, T high =220.

[0108] It should be further explained that, in addition to marking the candidate bright areas, isolated high-brightness noise points also need to be removed—the 8-neighborhood connectivity algorithm is used to perform connected component analysis on each candidate bright area; the number of pixels in each connected component of each candidate bright area is calculated; connected components with a pixel count less than a sixth threshold are removed from the corresponding candidate bright areas. For example, the sixth threshold can be 50 (corresponding to an actual length of approximately 2-3 cm). Since shallow cracks are mostly continuous lines, isolated points with a small number of pixels are mostly noise points, small-area isolated bright areas (such as random noise from the sensor) are removed.

[0109] Step 5: Calculate the gradient value of each highlight candidate region and delete highlight candidate regions whose gradient value is less than or equal to the second threshold.

[0110] The edges of shallow cracks appear blurred due to scattering (gradient changes are gentle), while the edges of scratches, borders, and other interference are clearer (gradient changes are drastic). Therefore, step 5, based on step 4, further filters out shallow cracks that meet the edge characteristics of each marked highlight candidate region by calculating the gradient value of the highlight candidate region, combined with the edge features of the shallow cracks. This includes the following steps: Step 5.1: Calculate the gradient value of the grayscale image.

[0111] The gradient values ​​of an image are calculated using the Sobel or Prewitt operators (which reflect the degree of change in pixel grayscale). The smaller the gradient value, the more blurred the edges.

[0112] Step 5.2: Set the gradient threshold.

[0113] Statistically analyze the gradient value distribution of shallow crack areas (usually ≤30), set the upper limit threshold G1 (e.g., G1=30), retain areas with gradient values ​​≤G1 in the highlight candidate areas, and exclude clear edge areas with gradient values ​​>G1 (e.g., scratch edge gradient is about 50~80, border edge gradient >100).

[0114] Step 6: Determine whether each of the remaining bright candidate areas simultaneously meets the screening criteria for hot plate fault latent area. If it does, mark each bright candidate area that meets the screening criteria as a hot spot fault latent area.

[0115] The morphology of shallow cracks is a thin, continuous line or band, which needs to be quantified and distinguished by morphological parameters. Step 6, based on step 5, further filters out the bright candidate regions that meet both core and edge features, selecting those that meet the linear / band-like features. These filtered bright candidate regions then satisfy the morphological characteristics of shallow cracks. The filtering conditions are threefold: Condition 1: The aspect ratio of the bright candidate region ≥ the third threshold; Condition 2: The circularity of the bright candidate region ≤ the fourth threshold; Condition 3: The number of broken pixels in the bright candidate region ≤ the total number of pixels in the bright candidate region × the fifth threshold.

[0116] Specifically, step 6 includes: Step 6.1: Calculate the morphological parameters of shallow cracks.

[0117] For the highlighted candidate regions selected in steps 4 and 5, calculate the following parameters: 1. Aspect Ratio. Calculate the "length / width" of the smallest bounding rectangle of the shallow crack region. Since the aspect ratio of shallow cracks is usually ≥5:1 (e.g., a crack that is 10cm long and 1cm wide corresponds to an aspect ratio of approximately 10:1 in the image), 5:1 can be used as the third threshold described in step 6.

[0118] 2. Circularity. Circularity = l 2 / (4π×S), where l is the perimeter of the shallow crack region and S is the area of ​​the shallow crack region. The closer the roundness is to 0, the closer it is to a linear shape (the roundness of shallow cracks is usually ≤0.2), while the roundness of circular or irregular blocky regions (such as stains, junction boxes) is >0.5. 0.2 can be used as the fourth threshold mentioned in step 6.

[0119] 3. Number of fractured pixels. The shallow crack region is simplified into a single-pixel-width skeleton using a skeleton extraction algorithm (such as the Zhan-Suen algorithm). If the skeleton segments are continuous and without obvious breaks (the number of fractured pixels ≤ 10% of the total length of the shallow crack), it meets the linear characteristic. 10% can be used as the fifth threshold mentioned in step 6. The total length of the shallow crack can be calculated using the pixel coordinates of the start and end points of the shallow crack in the grayscale image.

[0120] Step 7: Delete the regions where the gray value is less than the first threshold, the gradient value is greater than the second threshold, the aspect ratio is less than the third threshold, the roundness is greater than the fourth threshold, and the gray value variance is less than the seventh threshold.

[0121] The purpose of step 7 is to eliminate other non-crack interference features. These include: 1. Scratch interference.

[0122] The ultraviolet reflection of scratches is weaker than that of shallow cracks, and the edges are clear and the shape is short and thick. The gray value of the scratched area is usually 120~170 (lower than 180~255 for shallow cracks), which meets the lower limit of the gray value threshold. T low =180 can directly eliminate most scratches, as described in step 7, by deleting areas with grayscale values ​​< the first threshold. Furthermore, the physical traces formed by mechanical damage at the scratch edges are relatively sharp, with gradient values ​​> 50 (far higher than the ≤ 30 for shallow cracks). These are further excluded using a gradient threshold G1 = 30, as described in step 7, by deleting areas with gradient values ​​> the second threshold. Even further, the scratches are mostly short straight lines with an aspect ratio < 3 (e.g., length 5cm, width 2cm), which does not meet the ≥ 5:1 threshold for shallow cracks. Therefore, areas with an aspect ratio < the third threshold are deleted, as described in step 7.

[0123] 2. Stains interfere with Stains (such as dust, bird droppings, etc.) absorb ultraviolet light, appearing as low-grayscale areas (dark spots). The grayscale value of stained areas is typically <100 (far lower than the bright characteristics of shallow cracks), directly passing the lower limit of the grayscale threshold. T low =180 removed. In addition, the stains are mostly irregular blocks with a roundness of >0.5 and no continuous linear skeleton, which conflicts with the morphological parameters of shallow cracks. The stain interference can be further removed by using a roundness threshold, such as deleting areas with a roundness greater than the fourth threshold as described in step 7.

[0124] 3. Cell texture interference Due to the series-parallel structure of the battery cells, they may exhibit faint stripes, but the grayscale value fluctuation is small (±10), and there are no obvious bright areas. These areas are excluded through grayscale variance analysis (grayscale variance of crack areas > 30, texture areas < 10). The grayscale variance = 30 can be used as the seventh threshold mentioned in step 7, and areas with grayscale variance < the seventh threshold are deleted as described in step 7.

[0125] Step 8: Obtain the actual physical coordinates of each hot spot fault latency zone.

[0126] include Figure 5 The following steps are shown: Step 8.1: Identify the four corner points of the photovoltaic module frame in the grayscale image, and establish a two-dimensional coordinate system with the top left corner point as the origin.

[0127] The four corner points of the photovoltaic module are the intersections of the dark lines on the frame, possessing the dual characteristics of "grayscale abrupt changes + geometric vertices," requiring further precise positioning based on edge detection. This step uses corner point positioning based on edge intersections, specifically: (1) Based on the four border lines extracted in step 2.2.6, calculate their pairwise intersections (such as the intersection of the horizontal border and the vertical border) to obtain the coordinates of four candidate corner points. x 1, y 1) ( x 2, y 2), ( x 3, y 3) and ( x 4, y 4).

[0128] First, mark the horizontal border line parallel to the x-axis of the image as... L 1 and L 2. Mark the vertical border line parallel to the y-axis of the image as... L 3 and L 4.

[0129] Then, respectively the straight lines L 1~ L 4. Express it using the general expression for a straight line.

[0130] Next, based on the general expression for a straight line, calculate the intersection point of the horizontal and vertical lines. The intersection point at the top left corner of the border is... L 1 (the horizontal line above) and L The intersection of 3 (the vertical lines on the left) and the intersection at the top right corner of the border is... L 1 (the horizontal line above) and L The intersection of 4 (the vertical lines on the right) and the bottom left corner of the border is the intersection point. L 2 (the horizontal line below) and L The intersection of line 3 (the vertical line on the left) and the bottom right corner of the border is the intersection point. L 2 (the horizontal line below) and L 4. The intersection of the vertical lines on the right.

[0131] (2) Calculate the side length and interior angle of the quadrilateral. If it meets the characteristics of a rectangle (opposite sides are equal, interior angles are close to 90°, and the ratio of side length is consistent with the standard size ratio of the component, such as 1640mm:992mm≈1.65:1), then the candidate corner point is confirmed to be valid.

[0132] Having determined the four corner points of the border, the side lengths of each side of the border can be directly calculated based on the pixel coordinates of these four corner points, according to the straight lines. L 1~ L The general expression for 4 can be used to calculate the included angle between two adjacent sides.

[0133] It should be further explained that, since the intersection points obtained from edge detection may have pixel-level errors (such as being located in the middle of the border edge instead of at the vertex), it is necessary to improve accuracy through sub-pixel corner detection (such as sub-pixel optimization of Harris corners, Shi-Tomasi algorithm). Specifically, within a small area (such as 5×5 pixels) around the candidate corner point, a grayscale value change curve is fitted (the grayscale valleys of the dark border lines form a cross at the corner point), and quadratic curve interpolation is used to calculate the corner point coordinates accurate to 0.1 pixels, reducing the error in subsequent size conversion.

[0134] Furthermore, a two-dimensional coordinate system is established with the top left corner of the photovoltaic module frame as the origin (x-axis along the length of the module, y-axis along the width). Step 8.2: Based on the pixel coordinates of the origin in the grayscale image, obtain the starting pixel coordinates and ending pixel coordinates of each hot spot fault latency zone.

[0135] Step 8.3: Calculate the actual physical size of the pixels in the grayscale image based on the pixel coordinates of the origin in the grayscale image.

[0136] For example, if the horizontal frame length of a photovoltaic module is 1640mm, and the area between the top left and top right corners contains 3280 pixels, then 1 pixel = 0.5mm.

[0137] Step 8.4: Based on the starting pixel coordinates, ending pixel coordinates, and the physical size corresponding to the pixel, obtain the starting physical coordinates and ending physical coordinates of each strip candidate area on the surface of the photovoltaic module.

[0138] For example, if a crack in an image extends from pixel (100,200) to (300,200), and 1 pixel = 0.5 mm, then the actual location is a horizontal crack on the component surface at a position 50 mm to 150 mm in the x direction and 100 mm in the y direction.

[0139] It should be noted that for multi-component arrays, the component number corresponding to the current image (such as the component in the 5th column of the 3rd row) can be determined by the UAV's GPS positioning data, and the final output is "component number + local coordinates" (such as "component 3-5, X: 50-150mm, Y: 100mm").

[0140] Example 2: Corresponding to Example 1, this example provides a photovoltaic module hot spot fault early warning system based on intelligent drone field inspection, including: An image acquisition module is used to acquire ultraviolet images from above the photovoltaic module; The image processing module is used to convert ultraviolet images into grayscale images; The first data analysis and processing module is used to obtain the gray value of each pixel and mark the area with a gray value greater than or equal to the first threshold as a highlight candidate area. The second data analysis and processing module is used to calculate the gradient value of each highlight candidate region and delete highlight candidate regions whose gradient value is less than or equal to the second threshold. The third data analysis and processing module is used to determine whether each of the remaining bright candidate areas simultaneously meets conditions one to three, and to mark each bright candidate area that simultaneously meets conditions one to three as a hot spot fault latent area; wherein, condition one: the aspect ratio of the bright candidate area is ≥ the third threshold, condition two: the circularity of the bright candidate area is ≤ the fourth threshold, and condition three: the number of broken pixels in the bright candidate area is ≤ the total number of pixels in the bright candidate area × the fifth threshold. The physical coordinate acquisition module is used to acquire the actual physical coordinates of each hot spot fault latency zone; The early warning signal generation module is used to generate early warning signals for each hot spot fault latency zone based on the actual physical coordinates.

[0141] Furthermore, the physical coordinate acquisition module includes: The border corner point recognition unit is used to identify the four corner points of the photovoltaic module border in a grayscale image; The coordinate system establishment unit is used to establish a two-dimensional coordinate system with the top-left corner point as the origin; The first coordinate acquisition unit is used to acquire the starting pixel coordinates and ending pixel coordinates of each hot spot fault latency zone based on the pixel coordinates of the origin in the grayscale image. The pixel size calculation unit is used to obtain the starting pixel coordinates and ending pixel coordinates of each hot spot fault latency zone based on the pixel coordinates of the origin in the grayscale image. The second coordinate acquisition unit is used to obtain the starting physical coordinates and ending physical coordinates of each strip candidate area on the surface of the photovoltaic module based on the starting pixel coordinates, ending pixel coordinates and the physical size corresponding to the pixel.

[0142] Furthermore, the photovoltaic module hot spot fault early warning system also includes: The image distortion correction module is used to perform radial and tangential distortion correction on grayscale images using the distortion coefficients of the ultraviolet sensor. The three-dimensional coordinate acquisition module is used to acquire the three-dimensional coordinates of the ultraviolet sensor when it takes pictures, based on the GPS data recorded by the drone and the three-dimensional coordinate offset of the ultraviolet sensor from the center of gravity of the drone. The mapping relationship establishment module is used to establish a mapping relationship model between the pixel coordinate system of the grayscale image and the three-dimensional coordinate system of the photovoltaic module; the expression of the mapping relationship model is: ;in,( u , v () represents the actual pixel coordinates of a point on the surface of the photovoltaic module in the ultraviolet image. X ,Y , Z () represents the actual three-dimensional coordinates of a point on the surface of a photovoltaic module. X s , Y s , Z s () represents the three-dimensional coordinates when the ultraviolet sensor takes a picture. f Indicates the focal length of the ultraviolet sensor, ( u 0, v 0) represents the principal point coordinates of the ultraviolet sensor; The pixel coordinate generation module is used to input the actual three-dimensional coordinates of each point on the surface of the photovoltaic module, the three-dimensional coordinates when the ultraviolet sensor is taking pictures, the focal length of the ultraviolet sensor, and the principal point coordinates of the ultraviolet sensor into the mapping relationship model, and output the actual pixel coordinates of each point on the surface of the photovoltaic module in the grayscale image. The pixel coordinate correction module is used to correct the coordinates of each pixel in the grayscale image using the actual pixel coordinates. The conversion factor acquisition module is used to calculate the ratio between the actual physical size of the photovoltaic module frame and the pixel size of the photovoltaic module frame in the grayscale image, and obtain the size conversion factor. The pixel size correction module is used to correct the size of each pixel in the grayscale image according to the size conversion factor.

[0143] The image processing module is used to perform filtering and histogram equalization on grayscale images.

[0144] The connected component analysis module is used to perform connected component analysis on each highlighted candidate region using the 8-neighborhood connectivity algorithm. The pixel count calculation module is used to calculate the number of pixels in each connected region of each highlight candidate region; The fourth data analysis and processing module is used to remove connected components with a pixel count less than the sixth threshold from the corresponding highlight candidate region.

[0145] The fifth data analysis and processing module is used to delete regions with gray values ​​less than the first threshold, regions with gradient values ​​greater than the second threshold, regions with aspect ratios less than the third threshold, regions with roundness greater than the fourth threshold, and regions with gray value variance less than the seventh threshold.

[0146] It should be understood that the terms "system," "device," "unit," and / or "module" as used in this specification are a method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0147] As indicated in this specification and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0148] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0149] It should be noted that the structures, proportions, sizes, etc., illustrated in the accompanying drawings are merely for illustrative purposes to aid those skilled in the art and are not intended to limit the scope of the invention. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to size, without affecting the effectiveness and purpose of the invention, should still fall within the scope of the disclosed technical content. Furthermore, terms such as "upper," "lower," "left," "right," and "middle" used in this specification are merely for clarity and not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

Claims

1. A method for early warning of hot spot faults in photovoltaic modules based on intelligent unmanned aerial vehicle (UAV) field inspection, characterized in that, Includes the following steps: A drone equipped with an ultraviolet sensor collects ultraviolet images from above photovoltaic modules; Convert the ultraviolet image to a grayscale image by performing the following steps; Obtain the grayscale value of each pixel, and mark the regions with grayscale values ​​greater than or equal to the first threshold as highlight candidate regions; Calculate the gradient value of each highlight candidate region and delete highlight candidate regions whose gradient value is ≤ the second threshold. Determine whether each of the remaining highlight candidate regions simultaneously satisfies conditions one to three; where, condition one: the aspect ratio of the highlight candidate region is ≥ the third threshold, condition two: the circularity of the highlight candidate region is ≤ the fourth threshold, and condition three: the number of broken pixels in the highlight candidate region is ≤ the total number of pixels in the highlight candidate region × the fifth threshold. Each candidate region that simultaneously meets conditions one through three is marked as a hot spot fault latent region; Obtain the actual physical coordinates of each hot spot fault latency zone; Early warning signals for each hot spot fault latency zone are generated based on the actual physical coordinates.

2. The method for early warning of hot spot faults in photovoltaic modules based on intelligent unmanned aerial vehicle (UAV) field inspection according to claim 1, characterized in that, Obtain the actual physical coordinates of each hot spot fault latency zone, including the following steps: Identify the four corner points of the photovoltaic module's frame in the grayscale image, and establish a two-dimensional coordinate system with the top left corner point as the origin; Based on the pixel coordinates of the origin in the grayscale image, obtain the starting pixel coordinates and ending pixel coordinates of each hot spot fault latency zone; Calculate the actual physical size of the pixels in the grayscale image based on the pixel coordinates of the origin in the grayscale image; Based on the starting pixel coordinates, ending pixel coordinates, and the physical size corresponding to the pixel, obtain the starting and ending physical coordinates of each strip candidate area on the surface of the photovoltaic module.

3. The method for early warning of hot spot faults in photovoltaic modules based on intelligent unmanned aerial vehicle (UAV) field inspection according to claim 2, characterized in that, After converting the ultraviolet image to a grayscale image, the following steps are also included: Radial and tangential distortion corrections are performed on grayscale images using the distortion coefficients of an ultraviolet sensor. Based on the GPS data recorded by the drone during shooting and the three-dimensional coordinate offset of the ultraviolet sensor from the center of gravity of the drone, the three-dimensional coordinates of the ultraviolet sensor during shooting are obtained. Establish a mapping model between the pixel coordinate system of grayscale images and the three-dimensional coordinate system of photovoltaic modules; The actual three-dimensional coordinates of each point on the surface of the photovoltaic module, the three-dimensional coordinates captured by the ultraviolet sensor, the focal length of the ultraviolet sensor, and the principal point coordinates of the ultraviolet sensor are input into the mapping relationship model, and the actual pixel coordinates of each point on the surface of the photovoltaic module in the grayscale image are output; the coordinates of each pixel in the grayscale image are corrected using the actual pixel coordinates. The ratio between the actual physical size of the photovoltaic module frame and the pixel size of the photovoltaic module frame in the grayscale image is calculated to obtain the size conversion factor; the size of each pixel in the grayscale image is corrected according to the size conversion factor.

4. The method for early warning of hot spot faults in photovoltaic modules based on intelligent unmanned aerial vehicle (UAV) field inspection according to claim 3, characterized in that, The expression for the mapping relationship model is: ; in,( u , v () represents the actual pixel coordinates of a point on the surface of the photovoltaic module in the ultraviolet image. X , Y , Z () represents the actual three-dimensional coordinates of a point on the surface of a photovoltaic module. X s , Y s , Z s () represents the three-dimensional coordinates when the ultraviolet sensor takes a picture. f Indicates the focal length of the ultraviolet sensor, ( u 0, v 0) represents the coordinates of the principal point of the ultraviolet sensor.

5. A method for early warning of hot spot faults in photovoltaic modules based on intelligent unmanned aerial vehicle (UAV) field inspection, as described in claim 3 or 4, characterized in that, Before using the distortion coefficients of the ultraviolet sensor to perform radial and tangential distortion correction on the grayscale image, the following steps are also included: filtering and histogram equalization of the grayscale image.

6. A method for early warning of hot spot faults in photovoltaic modules based on intelligent unmanned aerial vehicle (UAV) field inspection, as described in any one of claims 1-4, characterized in that, Before calculating the gradient value of each highlight candidate region, the following steps are also included: performing connected component analysis on each highlight candidate region using the 8-neighborhood connectivity algorithm; calculating the number of pixels in each connected component of each highlight candidate region; and deleting connected components with a pixel count less than the sixth threshold from the corresponding highlight candidate region.

7. A method for early warning of hot spot faults in photovoltaic modules based on intelligent unmanned aerial vehicle (UAV) field inspection, as described in any one of claims 1-4, characterized in that, After determining whether each remaining candidate region of highlighting satisfies conditions one to three simultaneously, the following steps are also included: deleting regions with gray values ​​< the first threshold, regions with gradient values ​​> the second threshold, regions with aspect ratios < the third threshold, regions with roundness > the fourth threshold, and regions with gray value variance < the seventh threshold.

8. A photovoltaic module hot spot fault early warning system based on intelligent unmanned aerial vehicle (UAV) field inspection, characterized in that, include: An image acquisition module is used to acquire ultraviolet images from above the photovoltaic module; The image processing module is used to convert ultraviolet images into grayscale images; The first data analysis and processing module is used to obtain the gray value of each pixel and mark the area with a gray value greater than or equal to the first threshold as a highlight candidate area. The second data analysis and processing module is used to calculate the gradient value of each highlight candidate region and delete highlight candidate regions whose gradient value is less than or equal to the second threshold. The third data analysis and processing module is used to determine whether each of the remaining bright candidate areas simultaneously meets conditions one to three, and to mark each bright candidate area that simultaneously meets conditions one to three as a hot spot fault latent area. Among them, the conditions are:

1. The aspect ratio of the highlight candidate area is greater than or equal to the third threshold; 2. The circularity of the highlight candidate area is less than or equal to the fourth threshold; 3. The number of broken pixels in the highlight candidate area is less than or equal to the total number of pixels in the highlight candidate area × the fifth threshold. The physical coordinate acquisition module is used to acquire the actual physical coordinates of each hot spot fault latency zone; The early warning signal generation module is used to generate early warning signals for each hot spot fault latency zone based on the actual physical coordinates.

9. A photovoltaic module hot spot fault early warning system based on intelligent unmanned aerial vehicle (UAV) field inspection according to claim 8, characterized in that, The physical coordinate acquisition module includes: The border corner point recognition unit is used to identify the four corner points of the photovoltaic module border in a grayscale image; The coordinate system establishment unit is used to establish a two-dimensional coordinate system with the top-left corner point as the origin; The first coordinate acquisition unit is used to acquire the starting pixel coordinates and ending pixel coordinates of each hot spot fault latent area based on the pixel coordinates of the origin in the grayscale image. The pixel size calculation unit is used to obtain the starting pixel coordinates and ending pixel coordinates of each hot spot fault latency zone based on the pixel coordinates of the origin in the grayscale image. The second coordinate acquisition unit is used to obtain the starting physical coordinates and ending physical coordinates of each strip candidate area on the surface of the photovoltaic module based on the starting pixel coordinates, ending pixel coordinates and the physical size corresponding to the pixel.

10. A photovoltaic module hot spot fault early warning system based on intelligent UAV field inspection according to claim 9, characterized in that, Also includes: The image distortion correction module is used to perform radial and tangential distortion correction on grayscale images using the distortion coefficients of the ultraviolet sensor. The three-dimensional coordinate acquisition module is used to acquire the three-dimensional coordinates of the ultraviolet sensor when it takes pictures, based on the GPS data recorded by the drone and the three-dimensional coordinate offset of the ultraviolet sensor from the center of gravity of the drone. The mapping relationship establishment module is used to establish a mapping relationship model between the pixel coordinate system of the grayscale image and the three-dimensional coordinate system of the photovoltaic module; the expression of the mapping relationship model is: ; in,( u , v () represents the actual pixel coordinates of a point on the surface of the photovoltaic module in the ultraviolet image. X , Y , Z () represents the actual three-dimensional coordinates of a point on the surface of a photovoltaic module. X s , Y s , Z s () represents the three-dimensional coordinates when the ultraviolet sensor takes a picture. f Indicates the focal length of the ultraviolet sensor, ( u 0, v 0) represents the principal point coordinates of the ultraviolet sensor; The pixel coordinate generation module is used to input the actual three-dimensional coordinates of each point on the surface of the photovoltaic module, the three-dimensional coordinates when the ultraviolet sensor is taking pictures, the focal length of the ultraviolet sensor, and the principal point coordinates of the ultraviolet sensor into the mapping relationship model, and output the actual pixel coordinates of each point on the surface of the photovoltaic module in the grayscale image. The pixel coordinate correction module is used to correct the coordinates of each pixel in the grayscale image using the actual pixel coordinates. The conversion factor acquisition module is used to calculate the ratio between the actual physical size of the photovoltaic module frame and the pixel size of the photovoltaic module frame in the grayscale image, and obtain the size conversion factor. The pixel size correction module is used to correct the size of each pixel in the grayscale image according to the size conversion factor.

11. A photovoltaic module hot spot fault early warning system based on intelligent unmanned aerial vehicle (UAV) field inspection as described in claim 9 or 10, characterized in that, Also includes: The image processing module is used to perform filtering and histogram equalization on grayscale images.

12. A photovoltaic module hot spot fault early warning system based on intelligent unmanned aerial vehicle (UAV) field inspection as described in any one of claims 8-10, characterized in that, Also includes: The connected component analysis module is used to perform connected component analysis on each highlighted candidate region using the 8-neighborhood connectivity algorithm. The pixel count calculation module is used to calculate the number of pixels in each connected region of each highlight candidate region; The fourth data analysis and processing module is used to remove connected components with a pixel count less than the sixth threshold from the corresponding highlight candidate region.

13. A photovoltaic module hot spot fault early warning system based on intelligent unmanned aerial vehicle (UAV) field inspection as described in any one of claims 8-10, characterized in that, Also includes: The fifth data analysis and processing module is used to delete regions with gray values ​​less than the first threshold, regions with gradient values ​​greater than the second threshold, regions with aspect ratios less than the third threshold, regions with roundness greater than the fourth threshold, and regions with gray value variance less than the seventh threshold.

Images