Desert photovoltaic equipment inspection method and system based on unmanned aerial vehicle and AI algorithm

Abstract

The invention relates to the technical field of equipment inspection, in particular to a desert photovoltaic equipment inspection method and system based on an unmanned aerial vehicle and an AI algorithm. The method comprises the following steps: obtaining the local wind sand density of each pixel point in each inspection image according to the local hue feature of each pixel point in each inspection image and the local saturation distribution features of matched pixel points in different inspection images; according to the position distribution characteristics of edge pixel points on different edge lines in the inspection image, obtaining the equipment performance degree of each pixel point in the inspection image; obtaining a plurality of blocks of each inspection image based on a preset block side length; and according to the splicing probability distribution of the pixel points in each block in different training images, obtaining a splicing coefficient value corresponding to a preset block side length, and screening out an optimal block side length and an optimal spliced block forming an optimal spliced image. According to the method, the optimal spliced image which more comprehensively shows the equipment state during construction inspection is obtained, so that the accuracy of equipment inspection is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment inspection, and specifically relates to a method and system for inspecting desert photovoltaic equipment based on an unmanned aerial vehicle (UAV) and an AI algorithm. Background Art

[0002] Desert centralized photovoltaic power stations cover a large area, have a large number of equipment, and are located in harsh environments such as high temperature and sandstorms. Therefore, it is necessary to inspect the construction process of photovoltaic equipment.

[0003] In the prior art, the inspection area of the photovoltaic equipment construction process is photographed by a UAV, and edge recognition technology is used to obtain the shape of the photovoltaic equipment. However, due to the influence of sandstorms in different desert blocks, the degree of interference of the equipment blocks in different images by sandstorms is different. Directly applying edge recognition technology to the images results in blurred edges of the recognized facilities and poor recognition effect of the photovoltaic equipment. Summary of the Invention

[0004] In order to solve the technical problem of poor recognition effect of photovoltaic equipment without considering the influence of sandstorms, the purpose of the present invention is to provide a method and system for inspecting desert photovoltaic equipment based on a UAV and an AI algorithm. The specific technical solutions adopted are as follows: The present invention proposes a method for inspecting desert photovoltaic equipment based on a UAV and an AI algorithm, and the method includes: Obtain multiple inspection images of the inspection area of the photovoltaic equipment; obtain the matching pixel points between different inspection images, and obtain the local sandstorm density of each pixel point in each inspection image according to the local hue feature of each pixel point in each inspection image and the local saturation distribution feature of the matching pixel points in different corresponding inspection images; obtain the equipment expression degree of each pixel point in the inspection image according to the position distribution feature of the edge pixel points on different edge lines in the inspection image. Obtain the splicing probability of each pixel point in each inspection image according to the change trend of the equipment expression degree and the local sandstorm density of each pixel point in each inspection image relative to the corresponding matching pixel points in other inspection images. Obtain multiple blocks of each inspection image based on a preset block side length; obtain the splicing coefficient value corresponding to the preset block side length according to the splicing probability distribution of the pixel points in each block in different training images, and screen out the optimal block side length and the optimal splicing blocks that make up the best spliced image. Inspect the equipment according to the best spliced image.

[0005] Further, the method for obtaining the matching pixel points includes: Obtain the matching pixel points between different inspection images after grayscale processing based on the SIFT matching algorithm.

[0006] Further, the method for obtaining the local wind-sand density includes: Obtain the saturation mean value of all pixel points within the neighborhood range of each pixel point on each inspection image in the HSV color space as the local saturation; obtain the maximum and minimum values of the local saturation of the matching pixel points in different inspection images, and normalize the local saturation of each pixel point in each inspection image as the relative saturation feature of each pixel point in each inspection image; Obtain the fluctuation degree of the hue of all matching pixel points within the neighborhood window of each pixel point on the inspection image in the HSV color space as the local hue feature; According to the relative saturation feature and the local hue feature, obtain the local wind-sand density of each pixel point on each inspection image. Both the relative saturation feature and the local hue feature are negatively correlated with the local wind-sand density.

[0007] Further, the method for obtaining the equipment performance degree includes: Perform edge detection on the grayscale inspection image to obtain multiple edge lines corresponding to the inspection image; obtain the angle between the tangent line of each edge pixel point on each edge of each inspection image and the horizontal direction as the tangent angle of each edge pixel point; According to the change trend of the tangent angles of all edge pixel points on each edge line, obtain the equipment performance degree of each edge pixel point; For other pixel points except the edge pixel points, set the corresponding equipment performance degree to the positive integer 1.

[0008] Further, the step of obtaining the equipment performance degree of each edge pixel point according to the change trend of the tangent angles of all edge pixel points on each edge line includes: Construct a tangent angle broken line of all edge pixel points on each edge line in the order of pixel positions, and obtain the extreme points on the tangent angle broken line; obtain the upper envelope line fitted by all maximum points and the lower envelope line fitted by all minimum points, calculate the mean value of the data of all edge pixel points between the upper envelope line and the lower envelope line to form a new tangent angle broken line, and perform a preset number of processes on the tangent angle broken line to obtain a tangent correction angle broken line; Obtain the sum of the differences in the tangent correction angles between each edge pixel point on each edge of each inspection image and its adjacent edge pixel points as the equipment performance degree of each edge pixel point.

[0009] Further, the method for obtaining the splicing probability includes: Obtain the difference in the equipment performance degree between each pixel point on each inspection image and the matching pixel points in each other inspection image as the equipment performance difference; According to the device performance differences of each pixel point in each inspection image relative to the matching pixel points in different other inspection images, and the local sandstorm density of the corresponding pixel points in different inspection images, the splicing probability of each pixel point in each inspection image is obtained, and both the device performance difference and the local sandstorm density are negatively correlated with the splicing probability.

[0010] Further, the method for obtaining the splicing coefficient value includes: Obtain the average splicing probability of all pixel points within each block in each inspection image as the splicing probability level corresponding to each block; Select the block with the maximum splicing probability level among the blocks at each same position in all inspection images, and use the block at each position in the corresponding inspection image as the splicing block; Obtain the cumulative value of the splicing probability levels corresponding to the splicing blocks at all positions, and calculate the product of the cumulative value and the preset block side length as the splicing coefficient value corresponding to the preset block side length.

[0011] Further, the method for obtaining the optimal block side length and the splicing blocks constituting the best spliced image includes: Select the preset block side length with the largest splicing coefficient value among different preset block side lengths, and use the corresponding preset block side length as the optimal block side length; for the optimal block side length, obtain the splicing blocks corresponding to all positions in different inspection images as the optimal splicing blocks to form the best spliced image.

[0012] Further, the method for obtaining the neighborhood range includes: With each pixel point as the center, construct a range with an equal width as the neighborhood range of each pixel point.

[0013] The present invention also provides a desert photovoltaic device inspection system based on a drone and an AI algorithm, including a memory, a processor, and a computer program stored in the memory and operable on the processor. When the processor executes the computer program, the steps of any one of the above-mentioned desert photovoltaic device inspection methods based on a drone and an AI algorithm are implemented.

[0014] The present invention has the following beneficial effects: In view of the error impact caused by slight shaking of the drone, the present invention obtains the matching pixel points between different inspection images, quantifies the dust interference degree in the inspection environment according to the local hue characteristics of each pixel point in each inspection image and the local saturation distribution characteristics of the matching pixel points in different inspection images; obtains multiple edge lines of each inspection image, and obtains the device representation degree of each pixel point in the inspection image according to the position distribution characteristics of the edge pixel points on different edge lines in the inspection image, quantifying the characteristics of each pixel point representing the device; furthermore, obtains the splicing probability of each pixel point in each inspection image, which helps to select high-probability pixels to participate in splicing and reduce the influence of low-quality pixels such as blurring and repeated textures; obtains multiple blocks of each inspection image based on the preset block side length, facilitating the analysis of the sandstorm density and device feature distribution within the blocks; obtains the splicing coefficient value corresponding to the preset block side length according to the splicing probability distribution of the pixel points within each block in different training images, screens out the optimal block side length and the optimal splicing blocks that constitute the best spliced image, avoiding computational redundancy caused by too small blocks or loss of details due to too large blocks, and fuses multiple optimized block images; conducts inspections on the device. The present invention improves the accuracy of device inspection by obtaining the best spliced image that more comprehensively represents the device characteristics during the construction inspection of photovoltaic devices. Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0016] Figure 1 It is a flowchart of a method for inspecting desert photovoltaic devices based on a drone and an AI algorithm provided by an embodiment of the present invention; Figure 2 It is a flowchart of a method for obtaining the local sandstorm density provided by an embodiment of the present invention. Detailed Embodiments

[0017] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following, in combination with the accompanying drawings and preferred embodiments, details the specific embodiments, structures, features, and effects of a method and system for inspecting desert photovoltaic devices based on a drone and an AI algorithm proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this invention belongs.

[0019] The following specifically describes the specific solutions of a desert photovoltaic equipment inspection method and system based on drones and AI algorithms provided by the present invention in conjunction with the accompanying drawings.

[0020] Please refer to Figure 1 , which shows the method flow chart of a desert photovoltaic equipment inspection method provided by an embodiment of the present invention, specifically including: Step S1: Obtain multiple inspection images of the photovoltaic equipment inspection area.

[0021] In the embodiment of the present invention, in order to eliminate the shaking of the drone caused by environmental wind and the blurring of the collected images, it is necessary to obtain and analyze the original images during the construction process of the photovoltaic equipment; first, during the process of building solar photovoltaic panels in the desert, high-resolution equipment is used to collect images directly above the inspection block. During the collection process, the drone ensures the stability of its own height and position, which is more conducive to analysis. Obtain multiple inspection images of the photovoltaic equipment inspection area.

[0022] It should be noted that in the embodiment of the present invention, one image is collected every 5s, and a total of 12 images within 1 minute are collected for analysis; in other embodiments of the present invention, the acquisition time interval can be specifically set according to specific situations, and no limitation and elaboration will be made here.

[0023] Step S2: Obtain the matching pixel points between different inspection images, and obtain the local sandstorm density of each pixel point in each inspection image according to the local hue characteristics of each pixel point in each inspection image and the local saturation distribution characteristics of the matching pixel points in different inspection images corresponding to them; obtain multiple edge lines of each inspection image, and obtain the equipment performance degree of each pixel point in the inspection image according to the position distribution characteristics of the edge pixels on different edge lines in the inspection image.

[0024] During the construction process of the photovoltaic equipment, the positions of the pipe piles and brackets for the basic installation are ensured to be accurate. In order to eliminate the errors caused by the shaking of the drone, obtain the matching pixel points between different inspection images; it should be noted that in an embodiment of the present invention, the method for obtaining the matching pixel points includes: obtaining the matching pixel points between different inspection images after grayscale processing based on the SIFT matching algorithm. Among them, the inspection images after grayscale processing can eliminate the color channel differences and facilitate the processing of the images; the specific grayscale processing and SIFT matching algorithm are well-known technical means to those skilled in the art and will not be elaborated here.

[0025] Suspended particles in the wind and sand will scatter light, causing random offsets in the light path, reducing the purity of the color channels, and decreasing the overall saturation of the image; the sand and dust coverage makes the originally bright color patches present a dull effect, and the colors of the wind and sand in a local area are close to the same, with a lower hue complexity; therefore, by analyzing the changes in hue and saturation within the local range of each pixel point, it helps to understand the impact of wind and sand on the local range of pixel points in the image; based on the local hue characteristics of each pixel point in each inspection image and the local saturation distribution characteristics of the matching pixel points in different inspection images, the local wind and sand density of each pixel point in each inspection image is obtained.

[0026] Preferably, in an embodiment of the present invention, for the method of obtaining the local wind and sand density, please refer to Figure 2 , which shows a flowchart of a method for obtaining the local wind and sand density. It includes: Step S201: Obtain the saturation mean value of all pixel points within the neighborhood range of each pixel point on each inspection image in the HSV space as the local saturation; obtain the maximum and minimum values of the local saturation of the matching pixel points in different inspection images, and normalize the local saturation of each pixel point in each inspection image as the relative saturation feature of each pixel point in each inspection image.

[0027] By calculating the mean value, the saturation of all pixel points within the neighborhood range is analyzed as a whole to quantify the overall saturation level; during the continuous process of wind and sand, the sizes of the wind and sand in different inspection images may vary, showing different saturation situations. Therefore, by analyzing the local saturation of the corresponding matching pixel points in different inspection images, the influence degree of the wind and sand within the range is reflected. The greater the local saturation, the greater the relative saturation feature, and the greater the influence of the wind and sand.

[0028] It should be noted that the method of normalizing the local saturation of pixel points based on the maximum and minimum values of the local saturation is as follows: obtain the difference between the maximum and minimum values of the local saturation as the saturation range; obtain the difference between the local saturation of each pixel point and the minimum value of the local saturation, and calculate the ratio of the difference result to the saturation range, that is, normalize the local saturation of each pixel point; the specific means are well-known technical means to those skilled in the art and will not be elaborated here.

[0029] Step S202: Obtain the degree of fluctuation of the hue of all matching pixel points within the neighborhood window of each pixel point on the inspection image in the HSV space as the local hue feature.

[0030] The sand and dust coverage will cause the originally bright color patches to present a dull effect through physical occlusion or light scattering. The more chaotic the hue distribution, the less affected by the sand and dust coverage, and the more it shows the characteristics of the equipment.

[0031] It should be noted that, in an embodiment of the present invention, the degree of fluctuation is represented by calculating the variance. The larger the variance, the greater the degree of fluctuation; the smaller the variance, the smaller the degree of fluctuation. In other embodiments of the present invention, the degree of fluctuation can also be represented by calculating the range and standard deviation. The specific means are well-known technical means to those skilled in the art and will not be elaborated herein.

[0032] Step S203: According to the relative saturation feature and the local hue feature, obtain the local sandstorm density of each pixel point on each inspection image. Both the relative saturation feature and the local hue feature are negatively correlated with the local sandstorm density.

[0033] It should be noted that, in the embodiment of the present invention, the larger the relative saturation feature, the smaller the scattered light affected by the sandstorm, and the smaller the local sandstorm density, showing a negative correlation. The larger the local hue feature, the more obvious the hue performance within the divided block of the inspection image, and the smaller the possibility of being the color covered by the sandstorm, showing a negative correlation.

[0034] In an embodiment of the present invention, the formula for the local sandstorm density is expressed as: ; Wherein, represents the local sandstorm density of the th pixel point in the th inspection image; represents the variance of the hue in the HSV space of all pixel points within the neighborhood range of the th pixel point in the th inspection image, that is, the local hue feature; represents the relative saturation feature of the th pixel point in the th inspection image.

[0035] In the formula for the local sandstorm density, Adding 0.01 in

[0036] is to avoid the denominator of the formula being 0 and the formula being meaningless. The larger the relative saturation feature, the more likely it is that the influence of the sandstorm is smaller. The larger the local hue feature, the greater the change in hue, and the more color features are shown, and the smaller the influence of the sandstorm. Therefore, the larger the relative saturation feature and the larger the local hue feature, the smaller the local sandstorm density.

[0037] Due to the influence of sand and dust, the blocks shown in different images are disturbed by sand and dust to different degrees. The contrast of their edges decreases, and there are depressions at the edges, resulting in blurred edges of the recognition device. By analyzing the position distribution of edge pixel points, the device performance of the pixel points is reflected. According to the position distribution characteristics of edge pixel points on different edge lines in the inspection images, the device performance of each pixel point in the inspection images is obtained.

[0038] Preferably, in an embodiment of the present invention, the method for obtaining the device performance includes: Perform edge detection on the grayscale-processed inspection images to obtain multiple edge lines corresponding to the inspection images; obtain the angle between the tangent of each edge pixel point on each edge of each inspection image and the horizontal direction, as the tangent angle of each edge pixel point; According to the change trend of the tangent angles of all edge pixel points on each edge line, obtain the device performance of each edge pixel point; Preferably, in an embodiment of the present invention, according to the change trend of the tangent angles of all edge pixel points on each edge line, obtaining the device performance of each edge pixel point includes: Construct a tangent angle broken line of all edge pixel points on each edge line in the order of the positions of the pixel points, and obtain the extreme points on the tangent angle broken line; obtain the upper envelope line fitted by all maximum points and the lower envelope line fitted by all minimum points, calculate the mean value of the data of all edge pixel points between the upper envelope line and the lower envelope line, form a new tangent angle broken line, and perform a preset number of processes on the tangent angle broken line to obtain a tangent correction angle broken line; Obtain the sum of the differences in the tangent correction angles between each edge pixel point and its adjacent edge pixel points on each edge of each inspection image, as the device performance of each edge pixel point.

[0039] It should be noted that, in an embodiment of the present invention, for the tangent angle of each tangent pixel point, in order to perform a filtering operation on each pixel point to eliminate the influence of sand and dust, the fitting method is to obtain it by using cubic sample interpolation. In other embodiments of the present invention, existing fitting methods such as polynomial fitting can also be used. The specific means are well-known technical means to those skilled in the art and will not be elaborated here.

[0040] It should be noted that, in the embodiments of the present invention, the preset number of processes can be specifically set according to the specific situation. In an embodiment of the present invention, it is set to 3, and no further limitation and elaboration will be made here.

[0041] For other pixel points except edge pixel points, the corresponding device performance is set to the positive integer 1.

[0042] Step S3: Obtain the stitching probability of each pixel in each inspection image based on the change trends of the device performance and local sandstorm density of each pixel relative to the corresponding matching pixels in other inspection images.

[0043] There are differences in the feature representation capabilities of photovoltaic devices in different images. The position change range of each pixel in the images captured by the drone itself is relatively small. By analyzing the change trends of the device performance and local sandstorm density of each pixel in each inspection image relative to the corresponding matching pixels in other inspection images, the pixels that are more valuable for reflecting the device characteristics are identified; based on the change trends of the device performance and local sandstorm density of each pixel in each inspection image relative to the corresponding matching pixels in other inspection images, obtain the stitching probability of each pixel in each inspection image.

[0044] Preferably, in an embodiment of the present invention, the method for obtaining the stitching probability includes: Obtain the difference in device performance between each pixel in each inspection image and the matching pixels in each other inspection image as the device performance difference; Based on the device performance differences between each pixel in each inspection image and the matching pixels in different other inspection images, and the local sandstorm density of the corresponding pixels in different inspection images, obtain the stitching probability of each pixel in each inspection image. Both the device performance difference and the local sandstorm density are negatively correlated with the stitching probability.

[0045] In an embodiment of the present invention, the formula for the stitching probability is expressed as: ; Wherein, represents the stitching probability of the th pixel in the th inspection image; represents the device performance of the th pixel in the th inspection image; represents the local sandstorm density of the th pixel in the th inspection image; represents the device performance of the corresponding matching pixel of the th pixel in the th inspection image; represents the local sandstorm density of the corresponding matching pixel of the th pixel in the th inspection image; represents the number of other inspection images; represents the normalization function.

[0046] In the formula for the splicing probability, Adding 0.01 is to avoid the denominator of the formula being 0, which would make the formula meaningless; the smaller the local wind-sand density, the less affected by the wind-sand, and the higher the reference value; represents the th pixel point in the th inspection image relative to the th pixel point in the th inspection image, which is the difference in the device performance between the corresponding matching pixel points, that is, the device performance difference. The smaller the device performance difference, the closer the performance at the corresponding pixel point positions, and the more it can represent the actual image features, and the greater the splicing probability.

[0047] Step S4: Obtain multiple blocks for each inspection image based on the preset block side length; according to the splicing probability distribution of the pixel points in each block of different training images, obtain the splicing coefficient value corresponding to the preset block side length, and screen out the optimal block side length and the optimal splicing blocks that make up the best spliced image.

[0048] To avoid problems such as splitting and sharpening in the splicing result caused by using a single pixel point for splicing, local blocks are analyzed, and multiple blocks for each inspection image are obtained based on the preset block side length; it should be noted that in an embodiment of the present invention, due to the analysis of the preset block side length, the preset block is a square, and the common divisors of the length and width of the image are obtained, and the same values are selected in ascending order to obtain multiple preset block side lengths, and the image is divided. Among them, to ensure the analysis of a sufficient amount of data, the minimum preset block side length is 3. That is, if the length is 40, the common divisors are 1, 2, 4, 5, 10, 20, 40, and the width is 20, the common divisors are 1, 2, 4, 5, 10, 20, then the preset block side lengths are 4, 5, 10, 20. In other embodiments of the present invention, the size of the preset block side length can be specifically set according to the specific situation, which will not be limited and elaborated here.

[0049] Preferably, in an embodiment of the present invention, the method for obtaining the splicing coefficient value includes: Obtain the average splicing probability of all pixel points in each block of each inspection image as the splicing probability level corresponding to each block; Select the block with the largest splicing probability level among the blocks at the same position in all inspection images, and use the block at each position of the corresponding inspection image as the splicing block; Obtain the cumulative value of the splicing probability levels corresponding to the splicing blocks at all positions, and calculate the product of the cumulative value and the preset block side length as the splicing coefficient value corresponding to the preset block side length.

[0050] In an embodiment of the present invention, the formula for the splicing coefficient value is expressed as: ; Among them, represents the splicing coefficient value of the preset block side length ; represents the preset block side length; represents the maximum value of the splicing probability in the blocks at the th same position of all inspection images; represents that when the preset block side length is , the number of blocks on the inspection image.

[0051] In the formula of the splicing coefficient value, represents the cumulative value of the maximum values of the splicing probabilities in the blocks at each same position of all inspection images when the preset block side length is . The larger the maximum value of the corresponding splicing probability for each block at each position, the more accurate the performance characteristics, and the larger the splicing coefficient value.

[0052] The larger the splicing coefficient value under each preset block side length, the more device characteristics the spliced image shows under the preset block side length, which is more helpful for subsequent inspection of the device. Therefore, the larger the splicing coefficient value, the more necessary it is to perform division and splicing under the corresponding block side length to screen out the optimal block side length.

[0053] Preferably, in an embodiment of the present invention, the method for obtaining the optimal block side length and the splicing blocks constituting the best spliced image includes: Select the one with the largest splicing coefficient value among different preset block side lengths, and use the corresponding preset block side length as the optimal block side length; for the optimal block side length, obtain the splicing blocks corresponding to all positions in different inspection images as the optimal splicing blocks to form the best spliced image.

[0054] Step S5: Inspect the device according to the best spliced image.

[0055] The best spliced image can provide a high-precision and distortion-free device inspection image, which is helpful for more intuitively displaying the device status.

[0056] It should be noted that in another embodiment of the present invention, the equipment is inspected through the obtained optimal spliced image, including: performing Gaussian filtering on the optimal spliced image, inputting the processed optimal spliced image into a neural network model for training, judging the abnormal degree of equipment construction in the image through manual scoring of the image, and dividing it between [0, 1]; inputting different inspection areas of the photovoltaic equipment into the trained neural network model for analysis. When the abnormal degree is greater than the preset abnormal threshold, a reminder is sent to the staff in the construction site by the drone, and reminder data is sent to the staff at the control terminal; when the abnormal degree is less than or equal to the preset abnormal threshold, the drone moves horizontally to the next inspection area to be inspected, and at the same time, the relevant images collected are retained at the control terminal. Among them, the preset abnormal threshold is set to 0.3. In other embodiments of the present invention, the size of the preset abnormal threshold can be specifically set according to specific situations, and no limitation and elaboration are made here.

[0057] In summary, according to the local hue characteristics of each pixel point in each inspection image of the present invention, and the local saturation distribution characteristics of the matching pixel points in different inspection images, the local sandstorm density of each pixel point in each inspection image is obtained; according to the position distribution characteristics of the edge pixels on different edges in the inspection image, the equipment representation degree of each pixel point in the inspection image is obtained; multiple blocks of each inspection image are obtained based on the preset block side length; according to the splicing probability distribution of the pixel points in each block in different training images, the splicing coefficient value corresponding to the preset block side length is obtained, and the optimal block side length and the optimal splicing blocks constituting the optimal spliced image are selected. The present invention improves the accuracy of equipment inspection by obtaining the optimal spliced image that more comprehensively represents the equipment state during construction inspection.

[0058] The present invention also proposes a desert photovoltaic equipment inspection system based on a drone and an AI algorithm, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of any one of the desert photovoltaic equipment inspection methods based on a drone and an AI algorithm are implemented.

[0059] It should be noted that: the above sequence of embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0060] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the key points of each embodiment are the differences from other embodiments.

Claims

1. A desert photovoltaic equipment inspection method based on drones and AI algorithms, characterized in that: The method comprises: Acquire multiple inspection images of the inspection area of ​​the photovoltaic equipment; obtain matching pixel points between different inspection images, and obtain the local sand density of each pixel in each inspection image based on the local hue characteristics of each pixel in each inspection image and the local saturation distribution characteristics of the matching pixels in different inspection images; obtain the equipment representation of each pixel in the inspection image based on the position distribution characteristics of edge pixels on different edge lines in the inspection image; According to the change trend of the device representation degree and local sand density of each pixel point in each inspection image relative to the corresponding matching pixel points in other inspection images, the splicing probability of each pixel point in each inspection image is obtained; Based on the preset block side length, multiple blocks of each inspection image are obtained; according to the splicing probability distribution of the pixel points in each block in different training images, the splicing coefficient value corresponding to the preset block side length is obtained, and the optimal block side length and the optimal splicing block constituting the best splicing image are screened out; Inspect the equipment based on the best stitching image.

2. According to claim 1, a method for inspecting desert photovoltaic equipment based on drones and AI algorithms is characterized in that: The method for obtaining matching pixel points includes: The matching pixels between different inspection images after grayscale processing are obtained based on the SIFT matching algorithm.

3. According to claim 1, a method for inspecting desert photovoltaic equipment based on drones and AI algorithms is characterized in that: The method for obtaining the local wind sand density includes: The average saturation value of all pixels in the neighborhood of each pixel on each inspection image in the HSV space is obtained as the local saturation; the maximum and minimum local saturation values ​​of the matching pixels in different inspection images are obtained, and the local saturation of each pixel in each inspection image is normalized as the relative saturation feature of each pixel in each inspection image; The fluctuation degree of the hue of all matching pixels in the neighborhood window of each pixel on the inspection image in the HSV space is obtained as the local hue feature; The local wind-sand density of each pixel on each inspection image is obtained based on the relative saturation features and local hue features. Both the relative saturation features and the local hue features are negatively correlated with the local wind-sand density.

4. The method for inspecting desert photovoltaic equipment based on drones and AI algorithms according to claim 1 is characterized in that: The method for obtaining the device performance includes: Perform edge detection on the inspection image after grayscale processing to obtain multiple edge lines of the corresponding inspection image; obtain the angle between the tangent line of each edge pixel point on each edge of each inspection image and the horizontal direction as the tangent line angle of each edge pixel point; According to the change trend of the tangent angles of all edge pixels on each edge line, the device performance of each edge pixel is obtained; For pixels other than edge pixels, the corresponding device representation is set to a positive integer 1.

5. The method for inspecting desert photovoltaic equipment based on drones and AI algorithms according to claim 4 is characterized in that: The obtaining of the device performance of each edge pixel according to the change trend of the tangent angles of all edge pixels on each edge line includes: Construct the tangent angle polyline of all edge pixels on each edge line according to the position sequence of the pixel points, and obtain the extreme value points on the tangent angle polyline; obtain the upper envelope line fitted by all the maximum value points, and the lower envelope line fitted by all the minimum value points, calculate the mean of the data of all edge pixels between the upper envelope line and the lower envelope line, construct a new tangent angle polyline, perform a preset number of processing on the tangent angle polyline, and obtain the tangent correction angle polyline; The sum of the differences in tangent correction angles between each edge pixel and adjacent edge pixel points on each edge in each inspection image is obtained as the device representation of each edge pixel point.

6. The method for inspecting desert photovoltaic equipment based on drones and AI algorithms according to claim 1 is characterized in that: The method for obtaining the splicing probability includes: Obtain the difference in device representation between each pixel point in each inspection image and the matching pixel points in each other inspection image as the device representation difference; According to the equipment performance difference between each pixel point of each inspection image and the matching pixel points in other inspection images, as well as the local wind and sand density of the corresponding pixel points in different inspection images, the splicing probability of each pixel point in each inspection image is obtained. The equipment performance difference and the local wind and sand density are negatively correlated with the splicing probability.

7. The method for inspecting desert photovoltaic equipment based on drones and AI algorithms according to claim 1 is characterized in that: The method for obtaining the splicing coefficient value includes: Obtain the mean splicing probability of all pixels in each block in each inspection image as the splicing probability level corresponding to each block; Select the block with the highest probability of splicing among all the inspection images at the same position, and use the block at each position of the corresponding inspection image as the splicing block; The accumulated value of the splicing probability level corresponding to the splicing blocks at all positions is obtained, and the product between the accumulated value and the preset block side length is calculated as the splicing coefficient value corresponding to the preset block side length.

8. The method for inspecting desert photovoltaic equipment based on drones and AI algorithms according to claim 7 is characterized in that: The method for obtaining the optimal block side length and the spliced ​​blocks constituting the optimal spliced ​​image includes: The stitching coefficient value of different preset block side lengths is selected as the largest, and the corresponding preset block side length is used as the optimal block side length; for the optimal block side length, the stitching blocks corresponding to all positions in different inspection images are obtained as the optimal stitching blocks to form the best stitching image.

9. The method for inspecting desert photovoltaic equipment based on drones and AI algorithms according to claim 3 is characterized in that: The method for obtaining the neighborhood range includes: With each pixel as the center, a range of equal width is constructed as the neighborhood range of each pixel.

10. A desert photovoltaic equipment inspection system based on drones and AI algorithms, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of a desert photovoltaic equipment inspection method based on a drone and an AI algorithm as described in any one of claims 1 to 9 are implemented.

Images