Photovoltaic station multi-camera fusion inspection method, device, equipment and storage medium

Abstract

The present application discloses a multi-camera fusion inspection method, device, equipment and storage medium for photovoltaic stations, which relates to the field of computer vision technology, and discloses a multi-camera fusion inspection method for photovoltaic stations, including: obtaining an initial image set of a photovoltaic station taken by multiple cameras; calibrating the timestamp and spatial perspective of the initial image set to obtain a calibrated image set; performing image preprocessing according to the calibrated image set to obtain an enhanced image set; performing feature extraction and analysis according to the enhanced image set to obtain a fault image set, and completing the inspection of the photovoltaic station according to the fault image set. Through multi-camera data fusion, time synchronization, image preprocessing and feature extraction and analysis technical means, the automatic inspection and fault detection of photovoltaic stations under weather and environmental constraints are realized, and the accuracy and efficiency of inspection are improved.

Description

Technical Field

[0001] The present application relates to the field of computer vision technology, and in particular to a multi-camera fusion inspection method, device, equipment and storage medium for photovoltaic stations. Background Art

[0002] Photovoltaic power station equipment is exposed outdoors for a long time and is affected by environmental factors such as dust, temperature changes, wind, sand, rain and snow, which can easily cause material aging, surface cracks, loose connections and other faults. In order to ensure the stable operation of photovoltaic power stations, photovoltaic modules, inverters and other equipment need to be inspected regularly.

[0003] However, photovoltaic power stations are usually installed outdoors and exposed to natural light throughout the day. The light intensity varies with time and weather conditions, such as sunny days, cloudy days, sunrises, and sunsets, causing fluctuations in the quality of images collected by the camera. In the case of strong light reflection, low light, or shadow coverage, it is impossible to accurately capture the surface details of photovoltaic modules, making it difficult to identify tiny cracks or aging textures, and there is a risk of missed detection and false detection.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a photovoltaic site multi-camera fusion inspection, device, equipment and storage medium, aiming to solve the technical problem that in photovoltaic module monitoring, the image quality collected by the camera is affected by environmental factors such as lighting changes, reflections and glare, and there is a risk of missed detection and false detection.

[0006] To achieve the above purpose, the present application proposes a photovoltaic station multi-camera fusion inspection method, the method comprising:

[0007] Obtaining an initial set of images of the photovoltaic station field taken by multiple cameras;

[0008] Calibrate the timestamp and spatial perspective of the initial image set to obtain a calibrated image set;

[0009] Performing image preprocessing according to the calibration image set to obtain an enhanced image set;

[0010] Feature extraction and analysis are performed based on the enhanced image set to obtain a fault image set, and the inspection of the photovoltaic station is completed based on the fault image set.

[0011] In one embodiment, the step of performing feature extraction analysis based on the enhanced image set to obtain the fault image set includes:

[0012] Performing edge detection on the enhanced image set to obtain an edge detection image set;

[0013] Converting the pixel values ​​of the enhanced image set into a target temperature, and counting the size of the hot spot area according to a preset temperature and the target temperature, and obtaining a hot spot detection image set according to the statistical results;

[0014] Extracting texture features of the enhanced image set according to local binary patterns, and obtaining a texture analysis image set according to the texture features;

[0015] A fault image set is obtained according to at least one of the edge detection image, the hot spot detection image set and the texture analysis image set.

[0016] In one embodiment, the step of performing edge detection on the texture analysis image set to obtain an edge detection image set includes:

[0017] Calculating the gradient of an image in the texture analysis image set to obtain a gradient magnitude and a gradient direction of the image;

[0018] Obtaining a first threshold and a second threshold, and obtaining an edge candidate region according to the first threshold, the second threshold, the gradient amplitude, and the gradient direction;

[0019] Performing non-maximum suppression on the edge candidate region to obtain an initial edge image;

[0020] Edge connection and filtering are performed on the initial edge image to obtain an edge detection image set.

[0021] In one embodiment, the step of extracting texture features of the enhanced image set according to the local binary pattern and obtaining the texture analysis image set according to the texture features comprises:

[0022] Obtaining a central pixel of each image in the enhanced image set, and determining the number of pixels and the neighborhood radius of the local binary pattern according to the central pixel;

[0023] According to the number of pixels and the neighborhood radius, the grayscale value of each pixel is compared with the grayscale values ​​of its surrounding neighborhood pixels to obtain a comparison result;

[0024] Generate a binary texture value according to the comparison result, and obtain a texture feature map according to the binary texture value;

[0025] Detect abnormal areas based on texture changes in texture feature maps;

[0026] A texture analysis image set is generated according to the abnormal area.

[0027] In one embodiment, the image preprocessing includes noise reduction, adaptive contrast enhancement and brightness adjustment; the step of performing image preprocessing according to the calibration image set to obtain an enhanced image set includes:

[0028] Performing denoising on the calibration image set according to Gaussian filtering to obtain an initial denoised image set;

[0029] When it is detected that there is a fault image in the initial denoised image set, denoising the fault image according to bilateral filtering to obtain a target denoised image set;

[0030] Adaptively contrast enhancing the target denoised image set according to adaptive histogram equalization to obtain a contrast enhanced image set;

[0031] The brightness distribution of the contrast enhanced image set is analyzed to obtain an analysis result, and according to the analysis result, the highlight area in the contrast enhanced image set is compressed or the brightness of the dark area in the contrast enhanced image set is enhanced to obtain an enhanced image set.

[0032] In one embodiment, the step of performing adaptive contrast enhancement on the target denoised image set according to adaptive histogram equalization to obtain a contrast enhanced image set comprises:

[0033] Dividing each image in the target denoised image set into sub-blocks of a preset size according to adaptive histogram equalization;

[0034] Counting the number of pixels of the sub-block to obtain a histogram, calculating a pixel cumulative value of the histogram, and performing equalization according to the pixel cumulative value to obtain an equalized pixel value;

[0035] Adjusting the equalized pixel value according to a contrast limiting mechanism to obtain an enhanced local area;

[0036] The enhanced local regions are merged to obtain a contrast enhanced image set.

[0037] In one embodiment, the step of calibrating the timestamp and spatial perspective of the initial image set to obtain the calibrated image set includes:

[0038] Obtaining image timestamps and camera codes of images in the initial image set, and sorting the image timestamps according to the camera codes to obtain sorted timestamps;

[0039] Calibrate the image timestamp according to the sorting timestamp and the camera code to obtain a time-series synchronized image set;

[0040] constructing an integral image for the images in the time-series synchronous image set, and detecting feature points in the integral image according to a Hessian matrix;

[0041] Calculating distances between feature points in different integral images, and matching the feature points of the different integral images according to the distances to obtain initial matching points;

[0042] Filter invalid matching points from the initial matching points according to a ratio test to obtain target matching points;

[0043] A correction matrix is ​​calculated according to the target matching points, and normalization is performed according to the correction matrix to obtain a calibration image set.

[0044] In addition, to achieve the above-mentioned purpose, the present application also proposes a photovoltaic station multi-camera fusion inspection device, the photovoltaic station multi-camera fusion inspection device comprising: an acquisition module, used to acquire an initial image set of the photovoltaic station field taken by multiple cameras;

[0045] A calibration module, used to calibrate the timestamp and spatial viewing angle of the initial image set to obtain a calibrated image set;

[0046] A preprocessing module, used for performing image preprocessing according to the calibration image set to obtain an enhanced image set;

[0047] The analysis module is used to perform feature extraction analysis based on the enhanced image set to obtain a fault image set, and complete the inspection of the photovoltaic station based on the fault image set.

[0048] In addition, to achieve the above-mentioned purpose, the present application also proposes a multi-camera fusion inspection device for a photovoltaic station, the device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the multi-camera fusion inspection method for a photovoltaic station as described above.

[0049] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the multi-camera fusion inspection method of the photovoltaic station as described above are implemented.

[0050] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the photovoltaic station multi-camera fusion inspection method as described above.

[0051] One or more technical solutions proposed in this application have at least the following technical effects:

[0052] The invention adopts the technical means of obtaining an initial image set of a photovoltaic power station taken by multiple cameras; calibrating the timestamp and spatial perspective of the initial image set to obtain a calibrated image set; performing image preprocessing according to the calibrated image set to obtain an enhanced image set; performing feature extraction and analysis according to the enhanced image set to obtain a fault image set, and completing the inspection of the photovoltaic power station according to the fault image set. This solves the problem of difficulty in detecting defects such as tiny cracks, hot spots and material aging caused by illumination changes, perspective limitations and image quality problems in photovoltaic component monitoring. Compared with the existing technology, this invention realizes all-round real-time monitoring and accurate fault detection of photovoltaic power station equipment, improves the coverage, accuracy and efficiency of inspection, and at the same time enhances the adaptability under complex illumination and environmental conditions, ensuring a more reliable automated inspection effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0055] Figure 1 A flowchart diagram of the first embodiment of the photovoltaic station multi-camera fusion inspection method provided in this application;

[0056] Figure 2 A flow chart of the second embodiment of the photovoltaic station multi-camera fusion inspection method provided in this application;

[0057] Figure 3 A flowchart diagram of the third embodiment of the photovoltaic station multi-camera fusion inspection method of the present application;

[0058] Figure 4 A schematic diagram of crack detection provided in Example 3 of the photovoltaic station multi-camera fusion inspection method of this application;

[0059] Figure 5 A schematic diagram of a binary mask image provided in Embodiment 3 of the photovoltaic station multi-camera fusion inspection method of the present application;

[0060] Figure 6 This is a schematic diagram of the module structure of a multi-camera fusion inspection device for a photovoltaic station according to an embodiment of the present application;

[0061] Figure 7Schematic diagram of the equipment structure of the hardware operating environment involved in the photovoltaic station multi-camera fusion inspection method in the embodiment of the present application.

[0062] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0063] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0064] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0065] The main solution of the embodiment of the present application is: obtain the initial image set of the photovoltaic station field taken by multiple cameras; calibrate the timestamp and spatial perspective of the initial image set to obtain a calibrated image set; perform image preprocessing based on the calibrated image set to obtain an enhanced image set; perform feature extraction and analysis based on the enhanced image set to obtain a fault image set, and complete the inspection of the photovoltaic station field based on the fault image set. In this embodiment, for the convenience of description, the following is explained by identifying the multi-camera fusion inspection equipment of the photovoltaic station as the execution subject.

[0066] Since the image quality collected by the camera in the existing photovoltaic module monitoring is affected by environmental factors such as lighting changes, reflections and glare, there is a risk of missed detection and false detection. The present application provides a solution, which adopts the method of obtaining an initial image set of a photovoltaic station field taken by multiple cameras; calibrating the timestamp and spatial perspective of the initial image set to obtain a calibration image set; performing image preprocessing according to the calibration image set to obtain an enhanced image set; performing feature extraction and analysis according to the enhanced image set to obtain a fault image set, and completing the inspection of the photovoltaic station field according to the fault image set. The technical means solves the problem of difficulty in detecting defects such as tiny cracks, hot spots and material aging caused by lighting changes, perspective limitations and image quality problems in photovoltaic module monitoring. Compared with the existing technology, it realizes all-round real-time monitoring and accurate fault detection of photovoltaic power station equipment, improves the coverage, accuracy and efficiency of inspection, and enhances the adaptability under complex lighting and environmental conditions, ensuring a more reliable automated inspection effect.

[0067] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a photovoltaic station multi-camera fusion inspection device, etc. The following takes the photovoltaic station multi-camera fusion inspection device as an example to illustrate this embodiment and the following embodiments.

[0068] Based on this, the embodiment of the present application provides a photovoltaic station multi-camera fusion inspection method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the multi-camera fusion inspection method for photovoltaic stations of the present application.

[0069] In this embodiment, the photovoltaic station multi-camera fusion inspection method includes steps S10 to S40:

[0070] Step S10, obtaining an initial image set of the photovoltaic station field taken by multiple cameras;

[0071] It should be noted that a photovoltaic power station is a facility that uses solar photovoltaic modules to generate electricity. It is usually composed of multiple solar panels, inverters, junction boxes and related equipment. Photovoltaic power stations are widely distributed and often located outdoors. Photovoltaic power stations provide renewable energy to the power grid by converting sunlight into electrical energy. Since these devices are exposed to the natural environment for a long time, regular inspections are required to ensure their normal operation.

[0072] In addition, it should be noted that multi-camera refers to multiple cameras installed at different positions and angles of the photovoltaic station. These cameras work together to fully cover and detect various areas of the photovoltaic station. By collecting images from multiple perspectives, the limitations of a single camera perspective can be overcome and more detailed equipment information can be obtained.

[0073] Before the camera shoots, it is necessary to deploy multiple high-resolution cameras in key areas of the photovoltaic station. Key areas include but are not limited to photovoltaic modules, inverters, combiner boxes, etc. In this embodiment, it is necessary to determine the installation height of the camera and the field of view of the camera according to the height and width of the photovoltaic equipment. Usually, the camera needs to be installed at a higher position to overlook the entire photovoltaic module array, but it cannot be too far away from the photovoltaic equipment to avoid loss of resolution. The position of the camera needs to be debugged to make the camera resolution normal. Multiple cameras are not only reflected in quantity, but also in type. For detailed inspections, such as crack detection and hot spot detection, it is recommended to select cameras with at least 1080p or even 4K resolution. Since the camera is exposed to the outdoors for a long time, you can choose a device with IP66 or IP67 protection level to ensure that it is waterproof, dustproof, windproof, sandproof, high temperature and low temperature resistant, so as to maintain the effect of camera effect in extreme weather. In order to ensure the effectiveness of night monitoring, select a camera with infrared function to ensure that it can work normally in low light or at night. Infrared cameras can also be used to detect the temperature of electrical equipment such as inverters and combiner boxes to prevent equipment failures caused by excessive temperature. Choose the appropriate lens focal length according to the distance of different equipment such as photovoltaic modules, inverters, etc., and choose whether the camera has automatic zoom function according to the needs of the photovoltaic station, so as to flexibly adjust the monitoring range and details. In addition, the multi-camera can also be the camera mounted on the drone or inspection robot. The drone or inspection robot will take images of large-area photovoltaic modules from the air during the inspection. This and the fixed camera can obtain local images from different perspectives and can be used for subsequent image processing.

[0074] It should be noted that the initial image set refers to the raw image data collected from multiple cameras without any processing. This image set contains views of different devices and areas in the PV station and is the basis for subsequent processing. The initial image set is used to provide the basic data required for subsequent steps. By analyzing these raw images, the status of the PV panels can be detected.

[0075] Step S20, calibrating the timestamp and spatial viewing angle of the initial image set to obtain a calibrated image set;

[0076] It should be noted that the timestamp refers to the precise time information of each image when it is captured. Through the timestamp, the images taken by different cameras can be aligned in time sequence to ensure that the images captured in the same time period can be correctly matched, so as to analyze the equipment status at the same time. In an actual operation, cameras A, B, and C took images at 10:00, 10:01, and 10:02 respectively. If there is no alignment of the timestamps, their data will not be able to be compared at the same time point. Time calibration will obtain the data at 10:00 to ensure that the images obtained from A, B, and C represent the status of the photovoltaic equipment at the same time.

[0077] In addition, it should be noted that spatial perspective refers to the difference in perspective when different cameras shoot the same object from different positions and angles. These differences will cause the position, size and shape of the same object in different camera images to be different. Through parallax correction, the differences between the perspectives of different cameras can be eliminated, and the images from different cameras can be aligned in space to obtain a calibrated image with a unified perspective. In an actual operation, camera A shoots a photovoltaic component from the left, while camera B shoots from the front. The position and shape of the same component in the two images A and B will be significantly different. Perspective calibration aligns the two images to a unified reference perspective through mathematical transformation, so that the components are in the same position in the calibration image set.

[0078] It can be understood that the calibration image set, after being processed by time and space calibration, contains data from multiple cameras, ensuring that these images are aligned in time and space and can be used for subsequent analysis and detection. The calibration image set provides a unified data basis for subsequent feature extraction and fault analysis, ensuring that the data obtained from multiple cameras will not deviate due to time difference and perspective difference.

[0079] In a feasible implementation, step S20 may include steps S21 to S26:

[0080] Step S21, obtaining image timestamps and camera codes of images in the initial image set, and sorting the image timestamps according to the camera codes to obtain sorted timestamps;

[0081] It should be noted that the camera code is a unique identifier for each camera. Through this code, the system can distinguish data from different cameras.

[0082] It is understandable that the image timestamps of each camera are sorted to ensure that images in the same time period can be correctly matched. In a practical operation, cameras A and B capture images at 10:00:01 and 10:00:02 respectively. By extracting and sorting the timestamps, an ordered time series can be obtained, which is convenient for subsequent alignment.

[0083] Step S22, calibrating the image timestamp according to the sorting timestamp and the camera code to obtain a time-series synchronized image set;

[0084] It is understood that by sorting the timestamps, the images of different cameras are time-aligned. If there are slight timing differences, for example, the acquisition speed of a camera is different, time synchronization is performed through interpolation. Interpolation means that if the timestamp of an image of a camera is slightly earlier or later, the image of the corresponding time can be estimated through interpolation.

[0085] It should be noted that the time-series synchronized image set is formed by images from different cameras being synchronized in the same time window after time calibration.

[0086] Step S23, constructing an integral image for the images in the time-series synchronous image set, and detecting feature points in the integral image according to a Hessian matrix;

[0087] It is understandable that due to the different installation positions and angles of different cameras, there are parallaxes between images. Therefore, Speeded-Up Robust Features (SURF) can be used to correct the parallax of the image to ensure that the feature points in the same area are aligned in space. SURF detects and matches the feature points in the image and eliminates the parallax caused by different camera perspectives through the transformation matrix, so that the images of multiple cameras can be aligned in space. The geometric deformation and parallax caused by different camera perspectives are eliminated, so that the images of the same area can be correctly aligned under different cameras.

[0088] It should be noted that the integral image is not an image in the conventional sense, but a data structure used to accelerate the accumulation and calculation of pixels in the original image area. Each pixel value represents the cumulative sum of all pixels above and to the left of the point in the original image. Integral images are used to speed up the calculation of the pixel sum of a rectangular area, especially when detecting feature points. The integral image can be used to calculate the pixel accumulation of any area in a constant time. SURF uses integral images to accelerate the calculation of feature points. Integral images can quickly calculate the pixel sum in any rectangular area, making subsequent calculations more efficient.

[0089] In addition, it should be noted that the Hessian matrix is ​​a second-order derivative matrix that represents the curvature change of a local area in an image. By calculating the determinant of the Hessian matrix, the feature points of the image can be detected.

[0090] In addition, it can be understood that the key points in the image are detected by the determinant of the Hessian matrix. The Hessian matrix is ​​used to calculate the gradient information of each pixel, so as to find the most prominent areas of the image, such as corners and edges. The determinant of the Hessian matrix is ​​calculated for each pixel in the image. If the determinant value is large, the pixel is considered to be a feature point. The Hessian matrix is ​​defined as follows:

[0091]

[0092] Where, L xx , L yy , L xyIt is the second-order derivative of the image, which is used to describe the gradient changes of pixels in different directions.

[0093] Step S24, calculating the distance between the feature points in different integral images, and matching the feature points of the different integral images according to the distance to obtain initial matching points;

[0094] It can be understood that calculating the distance between feature points in different integral images can match feature points. By matching feature points, images can be spatially aligned. The smaller the distance, the more similar the two feature points are. They can be considered as feature points of the same object from different perspectives. Use the Euclidean distance to calculate the distance between each feature point descriptor and all feature point descriptors in the other image, find the most similar feature point in the other image for each feature point, and select the point with the smallest Euclidean distance as the matching point. The distance calculation formula is as follows:

[0095]

[0096] In the formula, p i and q i are the values ​​of the corresponding feature descriptors in the two images respectively.

[0097] It should be noted that the initial matching points are the first batch of feature point pairs obtained through distance calculation. These point pairs represent what the system believes may be corresponding points of the same object under different viewing angles.

[0098] Step S25, filtering invalid matching points from the initial matching points according to a ratio test to obtain target matching points;

[0099] It should be noted that the initial matching points are an unscreened set, which may contain false matching pairs, that is, different objects are mistakenly considered to be the same features. Therefore, further processing is required to improve the accuracy of matching. Ratio test is a method for evaluating the accuracy of feature point matching. By comparing the distance ratio of each feature point to the two closest feature points during the matching process, it can be determined whether these feature points are true matching pairs. The ratio test formula is as follows:

[0100]

[0101] Where d1 is the feature point with the smallest distance, and d2 is the feature point with the second smallest distance. Usually a threshold is set, such as 0.8, and the matching point is retained when the condition is met.

[0102] In addition, it should be noted that the valid feature point pairs screened out by the ratio test are called target matching points. They represent truly similar feature point pairs and are the basis for accurate image matching. Target matching points eliminate invalid matching pairs and provide accurate basic data for subsequent image fusion, correction and alignment. In the photovoltaic module image, the ratio test can screen out matching point pairs that truly represent the same crack, scratch or feature, thereby helping the system to align the image more accurately.

[0103] Step S26, calculating a correction matrix according to the target matching points, and normalizing according to the correction matrix to obtain a calibration image set.

[0104] It should be noted that the correction matrix is ​​a matrix that describes the geometric transformation of an image, which can map points in one image to corresponding points in another image. Through the correction matrix, the image of one camera can be "corrected" to the perspective of another camera, so that multiple images can be aligned in space and image fusion can be achieved under the same perspective. Image alignment can be performed through the homography matrix.

[0105] It can be understood that the homography matrix formula is as follows:

[0106]

[0107] Where H is a 3x3 transformation matrix, (x A ,y A ) is the pixel coordinate of the matching point in image A, (x B ,y B ) is the coordinate in the target image after transformation, and w is the normalization factor.

[0108] It should be noted that normalization refers to the conversion of images from one coordinate system to a standardized reference coordinate system through the application of a correction matrix to ensure that multi-camera images are represented at the same perspective. Normalization ensures the uniformity of the image, enables data from different perspectives to be seamlessly combined, and provides a stable foundation for subsequent analysis and detection.

[0109] In addition, it can be understood that the calculated (x B ,y B ,w) are homogeneous coordinates. In order to obtain the actual pixel position in image B, normalization is required. The normalization formula is as follows:

[0110]

[0111] In the formula, for each pixel point (x A ,y A ), use the homography matrix H to calculate the corresponding coordinates (x′) of the point in image BB ,y′ B ). The pixel value I of the point in image A is A (x A ,y A ) is mapped to the position (x′) of image B B ,y′ B ).

[0112] It is understandable that after the geometric transformation is completed, the image parallax caused by different camera positions and angles will be eliminated, and the spatial alignment effect between images will be achieved. The corrected images can be fused and analyzed in the same coordinate system. All geometrically transformed images are merged to generate a unified calibration image set. These images are aligned in time and space, and can provide consistent data input for subsequent processing and analysis.

[0113] Step S30, performing image preprocessing according to the calibration image set to obtain an enhanced image set;

[0114] It should be noted that image preprocessing refers to a series of processing operations on the original image to remove noise, enhance contrast, adjust brightness, etc., so as to improve the clarity and quality of the image. The enhanced image set is the image set obtained after image preprocessing. Compared with the original image, the enhanced image set has higher clarity and more obvious details.

[0115] Noise reduction is the use of filtering algorithms to remove random noise in images to ensure clarity, especially to reduce interference in low light or high reflectivity conditions. Gaussian filtering can smooth noise without significantly losing image details, and bilateral filtering can retain more edge information while removing noise.

[0116] Adaptive contrast enhancement can enhance the local contrast of an image, and is particularly suitable for contrast adjustment of the surface of a photovoltaic module under strong light or shadow conditions. This method can dynamically adjust the contrast of different areas in an image to improve the clarity of the overall image.

[0117] Brightness adjustment is based on the lighting conditions of the current scene, adaptively adjusting the image brightness. If the reflected light of the photovoltaic module causes overexposure, the intensity of the highlight area is reduced through brightness compression to ensure that the image is not covered by the overly bright area; if it is under low light conditions, the visibility of the dark area is improved through brightness enhancement.

[0118] Step S40: performing feature extraction and analysis based on the enhanced image set to obtain a fault image set, and completing the inspection of the photovoltaic station based on the fault image set.

[0119] It can be understood that feature extraction refers to the identification and extraction of valuable information for analysis from images, such as texture features, edge features, and temperature anomaly areas. These features can help identify the physical state and potential failures of photovoltaic modules.

[0120] It should be noted that the fault image set contains images in which problems are detected after feature extraction and analysis, and the photovoltaic modules in these images are marked with potential faulty or abnormal areas.

[0121] In addition, it can be understood that completing the inspection of the photovoltaic station based on the fault image set means summarizing the information in the fault image set, and an inspection report of the photovoltaic station can be generated based on the summarized information. The components that need to be inspected and maintained can be marked in the report to help operation and maintenance personnel quickly conduct on-site inspections.

[0122] This embodiment provides a multi-camera fusion inspection method for photovoltaic stations. The multi-camera fusion inspection technology solves the technical problems of low inspection efficiency, low accuracy and high cost caused by the limited viewing angle of a single camera, the influence of environmental factors and the long-term exposure of equipment to the outdoors in photovoltaic stations. By deploying multiple high-resolution cameras in key areas of photovoltaic stations, obtaining an initial image set, and then performing timestamp and spatial viewing angle calibration, image preprocessing, and feature extraction and analysis, a fault image set is finally generated, thereby achieving efficient and accurate inspection of photovoltaic stations. This method not only improves the coverage and detail capture capabilities of the inspection, but also enhances the detection accuracy of potential problems of photovoltaic modules, achieving the beneficial effects of reducing operation and maintenance costs and improving the reliability and safety of photovoltaic station operations.

[0123] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 2 The step S30 of the photovoltaic station multi-camera fusion inspection method includes steps S31 to S34:

[0124] Step S31, denoising the calibration image set according to Gaussian filtering to obtain an initial denoised image set;

[0125] It should be noted that during the inspection process, the image data collected by the camera is used as the input for noise reduction processing. These images may be affected by a variety of environmental factors, including low light, strong reflection, wind, sand or rain. These noises will interfere with the clarity of the image, making subsequent fault detection difficult to be accurate. The following problems may exist: Random noise, such as sensor noise, environmental interference, etc., causing the image to appear grainy or mottled. Strong light reflection or low light problems, high reflection on the surface of the photovoltaic panel or night / low light environment causes the image contrast to decrease and the details to be blurred.

[0126] It is understandable that Gaussian filtering is a commonly used smoothing filtering technology, which achieves the effect of noise reduction by convolving the image with a Gaussian function. The Gaussian filter assigns higher weights to the central area of ​​the image and lower weights to the area away from the center. This processing method can smooth the high-frequency noise in the image. In a specific implementation, a suitable Gaussian kernel size and standard deviation can be selected, and a suitable Gaussian filter kernel size (such as 3x3, 5x5, 7x7) can be selected. The larger the kernel, the stronger the smoothing effect and the better the denoising effect, but the image details may be more blurred. The standard deviation determines the degree of expansion of the Gaussian function. The larger the value, the stronger the smoothing effect. Usually, a suitable value can be selected according to the intensity of the image noise. The smoothed image after Gaussian filtering is output, and the noise is significantly reduced. The Gaussian kernel calculation formula is as follows 6:

[0127]

[0128] Where x and y are the pixel coordinates relative to the center of the kernel and y is the standard deviation.

[0129] It can be understood that the Gaussian kernel is convolved with the input image and applied to each pixel of the image to obtain a smoothed image. The basic steps of convolution are to slide the Gaussian kernel to each pixel of the image, perform a dot product between the kernel and the local area of ​​the image, and assign the result to the pixel. This can smooth the image and eliminate local noise. The convolution formula is as follows:

[0130]

[0131] Where I′(x,y) is the pixel value after Gaussian filtering, G(i,j) is the value of the Gaussian kernel, I(xi,yj) is the pixel value of the input image, and k is the radius of the kernel.

[0132] Step S32, when it is detected that there is a fault image in the initial denoised image set, denoising the fault image according to bilateral filtering to obtain a target denoised image set;

[0133] It is understandable that after the initial processing of Gaussian filtering, bilateral filtering can be used for further processing of areas where there may be faults or parts that require higher precision detection, such as microcracks or hot spots in photovoltaic modules, to retain more edge details and important texture information. This ensures that details are retained in local areas without sacrificing image accuracy.

[0134] It should be noted that in the initial denoised image set, some images showing abnormal features are detected, which are faulty images. These images may contain tiny cracks, hot spots, scratches, etc., indicating that the photovoltaic modules may have potential faults.

[0135] In addition, it should be noted that bilateral filtering is a nonlinear filter that considers not only the spatial distance but also the pixel value difference, so it can retain more edge details while denoising. The filter of bilateral filtering consists of two parts: spatial distance weight and pixel value difference weight. The spatial distance weight is generated by a Gaussian function, and the pixel value difference weight is also generated by a Gaussian function, but it acts on the pixel value difference. The calculation of the spatial distance weight is as follows 8:

[0136]

[0137] In the formula, the weight is calculated based on the spatial distance between pixels, i and j represent the positions of different pixels, and the farther the pixel is, the smaller the weight is. Spatial standard deviation σ r Controls the smoothness of the spatial range, similar to the standard deviation of Gaussian filtering, which determines the degree of smoothing.

[0138] The pixel value difference weight is calculated as follows:

[0139]

[0140] In the formula, i and j represent the positions of different pixels, and the intensity standard deviation σ s Determines the tolerance for pixel value differences. The larger the intensity standard deviation, the more obvious the smoothing effect, but more edge information will be lost. The smaller the intensity standard deviation, the more edge information will be retained, but the noise removal effect will be weaker.

[0141] The convolution kernel of bilateral filtering can be obtained from equations 8 and 9. The convolution kernel of bilateral filtering is calculated as follows:

[0142]

[0143] In the formula, G s (i,j) is the spatial Gaussian weight, G r (I(x,y),I(xi,yj)) is the weight of the pixel value difference, W p is the normalization constant, and B(x,y) is the pixel value after bilateral filtering.

[0144] It is understandable that if the environment detects strong light reflection, resulting in a strong highlight area on the surface of the photovoltaic panel, Gaussian filtering is not used, and bilateral filtering is used directly, which can retain the details of the photovoltaic panel surface while removing high-contrast noise.

[0145] In addition, it is understood that the output of the noise reduction process is an image in which the noise has been effectively eliminated after filtering, and the image clarity and detail are enhanced. The output image will be used in subsequent steps, and the noise-reduced image can be used for feature extraction processes such as edge detection and texture analysis to help identify potential problems such as cracks, hot spots, and equipment failures on photovoltaic modules. The noise-reduced image provides clearer input data, improving the accuracy of machine learning models or traditional detection algorithms, especially in complex environments such as strong light reflections or nighttime inspections.

[0146] Step S33, performing adaptive contrast enhancement on the target denoised image set according to adaptive histogram equalization to obtain a contrast enhanced image set;

[0147] It should be noted that, on the image after noise reduction, the contrast of the local area is enhanced, especially when there are shadows or strong light reflections on the surface of the photovoltaic module, to improve the image clarity and visibility of local details. Adaptive histogram equalization is a contrast enhancement method based on local areas rather than the entire image. It divides the image into several small areas, called sub-blocks, and performs histogram equalization on each sub-block, thereby enhancing the local contrast of the image.

[0148] It is understandable that in the actual application of PV station inspection, the surface of PV modules may be difficult to distinguish details due to dust, dirt and light reflection, especially in strong light and shadow areas. These factors can mask small fault features such as cracks, scratches and hot spots, affecting the accuracy of inspection. Through adaptive histogram equalization, the contrast of local areas of the image can be significantly enhanced, making the details in the image clearer. By improving the contrast of the image, small cracks, hot spots, scratches and other details on the surface of PV modules can be made more clearly visible, improving the accuracy of fault identification. In the shadows of PV modules or areas with uneven light distribution, the light can be balanced and the details can be enhanced, making faults in these areas easier to detect. By effectively improving the contrast of details, the inspection system can more accurately identify real faults and reduce false detections caused by noise.

[0149] In addition, it should be noted that the contrast enhanced image set is an image set that has been processed by adaptive histogram equalization. The details in these images have been significantly enhanced, especially showing more texture and detail features on the surface of photovoltaic modules.

[0150] In a feasible implementation, step S33 may include steps S331 to S334:

[0151] Step S331, dividing each image in the target denoised image set into sub-blocks of a preset size according to adaptive histogram equalization;

[0152] It should be noted that the preset size is the sub-block size selected according to the size of the image. For example, the image can be divided into 8x8 or 16x16 sub-blocks. The smaller the sub-block, the stronger the local contrast enhancement, but the computational complexity will also increase. According to the size of the image, calculate the number of sub-blocks after division. For example, for a 256×256 image, using a 16×16 sub-block size, it will be divided into 16×16 sub-blocks.

[0153] Step S332, counting the number of pixels of the sub-block to obtain a histogram, calculating a pixel cumulative value of the histogram, performing equalization according to the pixel cumulative value to obtain an equalized pixel value;

[0154] It should be noted that the histogram is a statistical graph of the number of pixels at each gray level in the image. The horizontal axis represents the gray levels 0 to 255, and the vertical axis represents the number of pixels at the corresponding gray level. The histogram of the sub-block is calculated by counting the pixels in each sub-block to generate a histogram, that is, the distribution of the number of pixels at each gray level (0-255). Each pixel in the sub-block is traversed, the number of pixels at each gray level 0-255 is counted, and the histogram of the sub-block is generated. The distribution of pixel gray values ​​is recorded in an array to form a histogram.

[0155] It should be noted that the pixel cumulative value CDF is calculated based on the histogram. The pixel cumulative value is the cumulative sum from the darkest gray level to the current gray level, which represents the total number of pixels below each gray level, that is, the number of pixels at each gray level is accumulated. The calculation of the cumulative distribution function is as follows:

[0156]

[0157] Where hist(j) is the number of pixels corresponding to gray level j. Assuming the gray value is i, it is the cumulative sum of all pixels from gray value 0 to i.

[0158] It can be understood that histogram equalization is to adjust the distribution of pixel values ​​to make the brightness values ​​more dispersed and evenly distributed, thereby improving the overall contrast of the image, making the bright parts brighter and the dark parts darker, thereby increasing visual clarity. In low-contrast images, many details may not be clearly presented due to insufficient brightness or contrast. Through equalization processing, the details in the image are enhanced, especially in images with uneven brightness or poor lighting, equalization can significantly improve the performance of image details. For example, in photovoltaic field inspections, histogram equalization can help enhance the surface details of photovoltaic modules, making cracks, hot spots and other problems easier to detect.

[0159] The original grayscale value of each pixel is mapped to a new pixel value, and the pixel value is adjusted according to the relative size of the CDF, so that the grayscale pixels that originally occupied a smaller number of pixels are expanded, and the grayscale pixels that originally occupied a larger number of pixels are compressed. The equalization formula is as follows:

[0160]

[0161] Where I′(x,y) is the pixel value after equalization, CFD min is the minimum cumulative distribution value, and N is the total number of pixels in the sub-block.

[0162] Step S333, adjusting the equalized pixel value according to the contrast limiting mechanism to obtain an enhanced local area;

[0163] In order to avoid noise amplification caused by excessive contrast enhancement, a contrast limiting mechanism can be used. When calculating the histogram of each sub-block, the contrast limiting mechanism sets an upper limit on the height of the histogram, that is, the number of pixels, to avoid local overexposure or darkness caused by a single gray level occupying too many pixels.

[0164] Usually, the contrast limiting mechanism controls the upper limit of the number of pixels at each gray level in the histogram by setting a clipping threshold. For example, if the number of pixels of a gray value in the histogram exceeds the clipping threshold, the excess part will be redistributed to other gray values ​​to ensure that there is no over-enhancement. If the clipping threshold is set to T, then if the number of pixels at a gray level exceeds T, the excess part will be clipped. The clipped pixels are redistributed to other gray levels to ensure that the cumulative distribution function (CDF) is smooth and avoid extreme brightness changes. Through clipping limitation, excessive enhancement of local contrast is prevented.

[0165] It should be noted that the enhanced local area is the image sub-block area adjusted by the contrast limitation mechanism, and the pixel value after equalization is appropriately limited, making the image more natural and the details more prominent without excessive enhancement. Limiting excessive enhancement allows the surface of the photovoltaic module to be processed evenly in both strong and weak light areas, reducing the problem of false detection caused by local magnification, and preventing image distortion caused by excessive enhancement, so that the inspection personnel can obtain more realistic component surface information when observing the image, and improve the reliability of the inspection results.

[0166] Step S334: merging the enhanced local regions to obtain a contrast enhanced image set.

[0167] It is understandable that merging refers to the process of stitching together all enhanced local areas obtained in the previous steps to form a complete image, but since the contrast enhancement of each sub-block is performed independently, there may be significant differences in brightness and contrast between different sub-blocks. Especially in areas with large changes in lighting conditions, unnatural boundaries will appear on edge transitions, reducing the visual effect of the image.

[0168] It should be noted that bilinear interpolation is used to interpolate pixels between adjacent sub-blocks, making the transition area smoother. The core idea of ​​bilinear interpolation is to generate pixel values ​​in the transition area through linear interpolation based on the pixel values ​​in adjacent sub-blocks, thereby visually making the edges of the sub-blocks smoother and more natural. It can ensure that each image achieves a natural transition based on local enhancement, and the overall visual effect is clearer and the details are richer. In the inspection of photovoltaic stations, this technology can effectively solve the problem of detail loss caused by uneven illumination, provide more reliable and accurate data support for the inspection system, and improve the accuracy and stability of the inspection results.

[0169] For each boundary pixel, bilinear interpolation needs to consider the pixel values ​​of its adjacent sub-blocks in both the horizontal and vertical directions. The basic formula for bilinear interpolation is to weight the pixel values ​​of adjacent sub-blocks according to the distance. The interpolation in the horizontal direction uses the upper and lower edge pixel values ​​of the adjacent sub-blocks, and performs linear interpolation according to the horizontal distance. The interpolation in the vertical direction uses the left and right edge pixel values ​​of the adjacent sub-blocks, and performs linear interpolation according to the vertical distance. The interpolation formula is as follows 13:

[0170]

[0171] In the formula, P(x, y) is the boundary pixel that needs to be interpolated, and the pixel values ​​of the four adjacent sub-blocks are Q 11 , Q 21 , Q 12 and Q 22 , x1, x2, y1 and y2 are the boundary coordinates of the sub-block respectively. x and y are the pixel coordinates that need to be interpolated.

[0172] In addition, it should be noted that the contrast enhanced image set is the final image set obtained after the merging process, and each image has completed the local contrast enhancement process, so that the overall image is more optimized in details and visual effects.

[0173] Step S34, analyzing the brightness distribution of the contrast enhanced image set to obtain an analysis result, and compressing the highlight area in the contrast enhanced image set or enhancing the brightness of the dark area in the contrast enhanced image set according to the analysis result to obtain an enhanced image set.

[0174] It should be noted that brightness adjustment is used to adaptively adjust the brightness of the image to adapt it to the current lighting conditions, ensuring that the image remains clearly visible in high light or low light conditions. Especially in the inspection of photovoltaic stations, due to the large changes in ambient light, the surface of photovoltaic modules may be overexposed or dark details may be lost. Through brightness adjustment, the brightness of the highlight area can be reduced and the details of the dark area can be enhanced. Brightness distribution analysis refers to the statistics of the brightness in the image, that is, the grayscale value, to understand the overall distribution of pixel brightness in the image. Through analysis, highlights, that is, areas with higher brightness, and dark areas, that is, areas with lower brightness, can be identified in the image.

[0175] By analyzing the global brightness distribution of the image, the distribution of brightness values ​​in the image is determined, and the lighting condition of the current image is determined to be overexposed or low light. If the input is a color image, it is first converted to a grayscale image to focus on the brightness distribution. The conversion formula is as follows 14:

[0176] I gary =0.2989×R+0.5870×G+0.1140×B (Formula 14)

[0177] In the formula, R, G, and B are the red, green, and blue channel values ​​of the image, respectively. gary is the corresponding grayscale value.

[0178] Based on the results of the global brightness distribution, the number of pixels for each grayscale value (0 to 255) is counted to generate a brightness histogram of the image. The horizontal axis of the histogram represents the brightness value, and the vertical axis represents the number of pixels corresponding to the brightness value. By observing the histogram, determine whether the brightness distribution of the image is concentrated in the dark area, indicating that the image is dark, or the highlight area indicates that the image is too bright. If most of the pixel brightness values ​​are concentrated in the higher area, it means that the image may be overexposed; if they are concentrated in the lower area, it means that the image is dark.

[0179] If the image's brightness histogram shows that a large number of pixels are concentrated in the bright area, such as grayscale values ​​close to 255, it means that the image is overexposed. If the brightness histogram shows that the pixels are mainly concentrated in the low-brightness area, such as grayscale values ​​close to 0, it means that the image is dark and the details are unclear.

[0180] Set brightness compression threshold: Set a high brightness compression threshold T high , for example T high =200. All pixel brightness greater than this value will be compressed. The pixels greater than the threshold are compressed as shown in equation 15:

[0181] I ′ (x,y)=T high +α×(I(x,y)-T high )(Formula 15)

[0182] In the formula, I ′ (x, y) is the compressed brightness value, I(x, y) is the original pixel brightness value, and α is the compression coefficient (usually 0<α<1), which is used to control the compression strength. The compressed brightness value is smaller than the original value, reducing the brightness of the highlight area.

[0183] Set the brightness enhancement threshold: Set a low brightness threshold T low , for example T low =50. All pixels with brightness below this value will be boosted. The brightness of pixels with brightness below the threshold is enhanced as shown in equation 16:

[0184] I ′ (x,y)=I(x,y)+β×(T low -I(x,y))(Equation 16)

[0185] In the formula, I ′ (x, y) is the enhanced brightness value, and β is the enhancement coefficient (usually 0<β<1), which is used to control the intensity of the enhancement.

[0186] In order to avoid over-enhancement or compression during brightness adjustment, it is usually necessary to readjust the dynamic range of the image so that the brightness value of the image can be more evenly distributed in the grayscale range of 0 to 255. The brightness of the image is remapped to the full grayscale range using a linear mapping formula to ensure that the brightness of the adjusted image does not exceed the range of 0-255. The calculation formula for controlling the brightness value is as follows:

[0187]

[0188] In the formula, I min and I max are the minimum and maximum brightness values ​​of the adjusted image, and I″(x,y) is the final brightness value.

[0189] During the inspection of photovoltaic sites, brightness adjustment can effectively reduce the detection blind spots caused by light differences and improve the accuracy and efficiency of the overall inspection, thereby ensuring that potential faults of photovoltaic modules can be discovered and handled in a timely manner.

[0190] This embodiment provides a multi-camera fusion inspection method for photovoltaic stations. Through the combined noise reduction technology of Gaussian filtering and bilateral filtering, combined with the method of adaptive histogram equalization and brightness adjustment, the common noise interference, uneven contrast and image quality degradation caused by changes in lighting conditions in photovoltaic station image acquisition are solved. Through these technical means, random noise and high-contrast noise in the image are effectively eliminated, edge details and important texture information are retained, the contrast of local areas is enhanced, and different lighting conditions are adapted, making the potential faults and abnormal areas of photovoltaic modules more obvious and easy to detect. This not only improves the accuracy of fault detection and the reliability of image analysis, but also improves the efficiency and effect of photovoltaic station inspections, ensuring the stable operation of photovoltaic power generation systems and reducing maintenance costs.

[0191] Based on the first embodiment of the present application, in the third embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 3 The step S40 of the photovoltaic station multi-camera fusion inspection method includes steps S41 to S44:

[0192] Step S41, performing edge detection on the enhanced image set to obtain an edge detection image set;

[0193] It should be noted that edge detection is an image processing technology that finds the boundary or outline of an object by detecting significant changes in the grayscale value in the image. Common edge detection algorithms include Sobel operator, Canny edge detection, Laplacian operator, etc. In this embodiment, the Canny edge detection algorithm is used to detect edge features in the image. Canny edge detection can accurately identify cracks, structural defects and other abnormalities in photovoltaic panels. Canny edge detection can capture clear edges more stably and is suitable for the detection of small cracks in photovoltaic modules.

[0194] In addition, it should be noted that the edge detection image set is an image set processed by an edge detection algorithm, which more accurately represents the contour and edge features in the enhanced image.

[0195] Please refer to Figure 4 , Figure 4 This is a schematic diagram of crack detection in the third embodiment of the photovoltaic station multi-camera fusion inspection method of the present application.

[0196] like Figure 4 As shown in the figure, after enhanced image processing and edge detection, the edge of the crack becomes very clear and sharp, forming a clear contrast with the surrounding normal area. The outline of the crack appears prominent in the image, which is convenient for quick identification.

[0197] In a feasible implementation, step S41 may include steps S411 to S414:

[0198] Step S411, calculating the gradient of the image in the texture analysis image set to obtain the gradient amplitude and gradient direction of the image;

[0199] It should be noted that the gradient of an image represents the rate and direction of change of the grayscale value. The gradient reflects which area of ​​the image has the fastest grayscale value change, which usually corresponds to the edge or texture features of the object. The gradient amplitude represents the intensity of the grayscale change, that is, how fast the rate of change is. The larger the amplitude, the more drastic the change, which usually occurs at the edge of the image or where the texture is prominent. The gradient direction represents the direction of the grayscale value change, that is, the gradient vector points to the direction of the fastest change. Direction information can help determine the direction of the edge.

[0200] Before calculating the gradient magnitude and gradient direction, the Canny algorithm will further smooth the image through Gaussian filtering internally to remove tiny noise to ensure the stability of edge detection.

[0201] The formula for convolution using Gaussian kernel is as follows:

[0202]

[0203] In the formula, K Gaussian is the Gaussian kernel, and I(x,y) is the input image pixel.

[0204] After further noise reduction, the horizontal convolution kernel and the vertical convolution kernel are obtained according to the Sobel operator convolution kernel, and convolution is performed again according to the calculated convolution kernel.

[0205] The horizontal convolution kernel K(x) is as follows:

[0206]

[0207] The vertical convolution kernel K(y) is as follows:

[0208]

[0209] Apply the above convolution kernel to each pixel in the image and calculate the gradient in the horizontal and vertical directions.

[0210] The calculation of the horizontal gradient is as follows:

[0211]

[0212] The vertical gradient is calculated as follows:

[0213]

[0214] In equations 21 and 22, I(x+i,y+i) is the pixel value of the input image, K x and K y They are the horizontal and vertical convolution kernels of the Sobel operator respectively.

[0215] Based on the horizontal direction G of Equation 21 and Equation 22 x and vertical direction G y The gradient of Calculate the gradient magnitude and gradient direction of each pixel.

[0216] The gradient amplitude is calculated as follows:

[0217]

[0218] The gradient direction is calculated as follows:

[0219]

[0220] In formulas 23 and 24, G x (x, y) is the horizontal gradient, G x (x,y) is the vertical gradient.

[0221] Step S412, obtaining a first threshold and a second threshold, and obtaining an edge candidate region according to the first threshold, the second threshold, the gradient amplitude, and the gradient direction;

[0222] It should be noted that the first threshold and the second threshold are the high threshold and the low threshold of Canny edge detection, respectively. Through these two thresholds, Canny edge detection can distinguish strong edges, weak edges and non-edges. Strong edge means that pixels with gradient amplitude greater than the high threshold are retained as strong edges. Weak edge means that pixels with gradient amplitude between the high threshold and the low threshold are marked as weak edges. Non-edge means that pixels with gradient amplitude lower than the low threshold are directly suppressed as non-edges. In actual situations, weak edges may represent small cracks on photovoltaic modules. If these weak edges are connected to strong edges, they may be real edges, and the Canny algorithm will retain them. Isolated weak edges, such as pseudo edges caused by noise, will be suppressed. For example, the low threshold can be set to 30 and the high threshold to 80, which means that pixels with gradient amplitude greater than 80 will be directly considered as strong edges, while pixels between 30 and 80 will be retained depending on the situation.

[0223] Step S413, performing non-maximum suppression on the edge candidate region to obtain an initial edge image;

[0224] It should be noted that non-maximum suppression is an image processing technique used to refine edges. During the processing, the algorithm searches along the gradient direction and sets pixels that are not local maxima to zero, ensuring that only the most significant edge pixels are retained in the final image. In order to accurately capture the edge, only the pixels with the largest gradient amplitude are retained, and the pixels with non-maximum values ​​are removed to ensure that the edge is thin and clear. For each pixel, check the two adjacent pixels in the gradient direction. If the gradient amplitude of the current pixel is not the local maximum, it is suppressed, that is, set to 0.

[0225] Step S414: edge connection and filtering are performed on the initial edge image to obtain an edge detection image set.

[0226] It should be noted that edge connection is an image processing technology that connects adjacent edge segments with the same direction to splice broken edge lines together to form a continuous edge. Edge filtering is used to remove unnecessary and erroneous edge information in an image. These erroneous edges may be caused by noise, shadows or other interference factors and need to be eliminated through further analysis.

[0227] It is understood that weak edges and strong edges can be connected to eliminate isolated weak edges caused by noise while retaining the actual edges. For each weak edge pixel, check whether there is a strong edge pixel among its neighboring pixels. If the weak edge pixel is connected to a strong edge, the weak edge is retained, otherwise it is suppressed. Small cracks on photovoltaic modules are usually composed of thin and relatively weak edges. By connecting the edges, it can be ensured that these crack edges can be continuously identified and tracked, providing accurate detection results.

[0228] By edge-connecting and filtering the initial edge image, it is ensured that all real edge lines can be displayed continuously and clearly, while effectively removing noise and pseudo-edge information. In the application of photovoltaic field inspection, this technology can help the system accurately identify linear defects on the surface of photovoltaic modules, such as cracks and scratches, and improve the accuracy and reliability of inspection. Finally, a binary image containing precise edges is output. The edges in the image are processed by non-maximum suppression and smoothing, and the edges are more refined.

[0229] Step S42, converting the pixel values ​​of the enhanced image set into a target temperature, and counting the size of the hot spot area according to a preset temperature and the target temperature, and obtaining a hot spot detection image set according to the statistical results;

[0230] It is understandable that hot spots may be caused by local damage to components, reduced efficiency or electrical failure. Accurately identifying hot spots can help to detect and repair problems in a timely manner to avoid further damage.

[0231] It should be noted that, according to the principle of infrared cameras, each pixel value in the image has a corresponding relationship with the actual temperature, and the grayscale value or color value in the image can be converted into corresponding temperature data, which is the target temperature. The preset temperature represents a temperature threshold, and the area above the threshold is marked as a possible hot spot area. The hot spot area is an abnormal temperature area identified during the temperature detection process, showing a higher temperature, which usually indicates that there is local heating on the surface of the photovoltaic module. The hot spot detection image set is a hot spot area identified by temperature analysis, which is marked and displayed in the form of an image. The location and size of the hot spot on the surface of the photovoltaic module are clearly marked in each image.

[0232] The image generated by the infrared camera is different from the ordinary RGB image. It records the temperature data of the object surface, so it is necessary to map the color or grayscale value in the image to the corresponding temperature value. Each pixel value (usually grayscale value or color value) of the infrared image corresponds to a temperature value. Infrared cameras generally have a fixed mapping relationship to convert pixel values ​​into temperature values.

[0233] Use the temperature mapping function provided by the camera to convert the pixel value in the image into the actual temperature. The conversion formula is as follows:

[0234] T(x,y)=α×(I(x,y))+b(Formula 25)

[0235] Where T(x,y) is the temperature value corresponding to the pixel, I(x,y) is the pixel value in the image, α is the proportional coefficient of temperature calibration, which determines the temperature change amplitude corresponding to each pixel value, and b is the offset coefficient, which is used to calibrate the starting temperature value.

[0236] According to the operating temperature range of the photovoltaic module, set a suitable temperature threshold T threshold . Usually, the temperature of a hot spot is significantly higher than the normal operating temperature. For example, setting the threshold to 80°C means that the area above 80°C may be a hot spot. Perform threshold detection on each pixel of the infrared image and mark the area above the threshold as a hot spot. Set the pixels exceeding the temperature threshold to white, that is, the hot spot area, and the rest to black, that is, the non-hot spot area. In this way, a binary mask image can be obtained, indicating the area where the fault may exist. Adjacent hot spot areas can be connected by dilation operation to ensure that the edges of the hot spots are intact. Use corrosion operation to remove isolated noise points in the image and retain the actual hot spot area. Count the size, location and number of hot spot areas. Based on the size and distribution of hot spots, it can be further inferred whether there is an abnormality in the photovoltaic module. For example, a large area of ​​hot spots may indicate serious damage to the photovoltaic module.

[0237] Please refer to Figure 5 , Figure 5This is a schematic diagram of a binary mask image of the third embodiment of the photovoltaic station multi-camera fusion inspection method of the present application.

[0238] like Figure 5 As shown in the figure, each pixel in the binary mask image is only black and white, representing different areas, the left side represents the non-hot spot area, and the right side represents the hot spot area. This clear binary segmentation can clearly highlight the target area in the photovoltaic module, such as cracks, hot spots, texture anomalies, etc., compared with the background.

[0239] The binary mask can also realize automatic fault alarm. When the number, area or temperature of hot spots detected in the mask exceeds a certain threshold, the system can automatically generate an alarm signal to prompt the inspection personnel to immediately pay attention to the operating status of certain photovoltaic modules.

[0240] Step S43, extracting texture features of the enhanced image set according to the local binary pattern, and obtaining a texture analysis image set according to the texture features;

[0241] It should be noted that the local binary pattern is a texture feature extraction method that generates a binary pattern by comparing the grayscale value of each pixel with its neighboring pixels, reflecting the texture characteristics of the local area. The texture analysis image set is an image set generated by applying LBP to the enhanced image set. The images in the texture analysis image set show the texture characteristics of the surface of the photovoltaic module, including normal and abnormal texture distribution.

[0242] In a feasible implementation, step S43 may include steps S431 to S434:

[0243] Step S431, obtaining the central pixel of each image in the enhanced image set, and determining the number of pixels and the neighborhood radius of the local binary pattern according to the central pixel;

[0244] It should be noted that when calculating the local binary pattern (LBP), the central pixel of the selected area is used as a reference and compared with the neighboring pixels around it. The central pixel determines the texture feature analysis of the entire local area. In LBP, the number of pixels represents the number of neighborhood pixels around the central pixel, usually taking values ​​such as 8, 16, etc. These pixels are used to compare with the central pixel to generate a binary pattern. The neighborhood radius represents the distance from the central pixel, that is, the interval between the central pixel and the neighboring pixels. The size of the radius determines the range of the neighborhood, which can be 1, 2 or a larger pixel value. The number of pixels and the neighborhood radius determine the accuracy and range of the LBP algorithm to capture image details. By adjusting these parameters, subtle textures in the image, such as cracks, scratches, etc., can be better identified.

[0245] Step S432, comparing the grayscale value of each pixel with the grayscale values ​​of its surrounding neighborhood pixels according to the number of pixels and the neighborhood radius to obtain a comparison result;

[0246] It should be noted that grayscale comparison refers to comparing the grayscale value of the central pixel with the grayscale value of each pixel in its neighborhood. If the grayscale value of the neighborhood pixel is greater than or equal to the grayscale value of the central pixel, it is marked as 1, otherwise it is marked as 0. Through grayscale comparison, a binary comparison result can be generated, for example: 10110010. This result represents the local texture characteristics of the central pixel relative to the neighborhood pixels.

[0247] Step S433, generating a binary texture value according to the comparison result, and obtaining a texture feature map according to the binary texture value;

[0248] It should be noted that the binary texture value is a binary number formed by combining the comparison results. For example, if the comparison result of 8 neighboring pixels is 10110010, this binary number is the texture value of the central pixel. The texture feature map generates a texture feature map by visualizing the binary texture value of each pixel in the image. This map can show the local texture distribution of the entire image and highlight the texture characteristics of the surface of the photovoltaic module, including cracks, scratches, aging areas, etc.

[0249] The grayscale comparison results are used to generate binary codes of local binary patterns, which can reflect the texture characteristics of local areas in the image. This process can effectively detect tiny cracks, scratches and other texture changes on the surface of photovoltaic modules.

[0250] Compare the central pixel I(x,y) with each neighboring pixel I p The comparison formula is as follows:

[0251]

[0252] Where P is the number of pixels around the center pixel, I(x,y) is the pixel value, and s(x) is a sign function, as shown in Equation 27:

[0253]

[0254] The result of each neighborhood pixel is encoded as a binary value, and the P comparison results are connected to form a binary pattern (i.e., LBP value), which reflects the local texture characteristics of the pixel. By performing this operation on the pixels of the entire image, the LBP pattern of each pixel is generated to form an LBP feature map.

[0255] Step S434, detecting abnormal areas according to texture changes in the texture feature map;

[0256] It is understandable that the texture distribution on the surface of a normal photovoltaic module is relatively uniform, and the changes in the LBP pattern are relatively smooth. The texture of defective areas or aged areas will show irregular changes, such as scratches, cracks, etc., and the distribution of the LBP pattern will be significantly different from that of normal areas. Therefore, in the texture feature map, areas that show discontinuous, irregular, or different texture patterns from the surrounding areas usually represent cracks, scratches, aging or other defects on the surface of the module.

[0257] Step S435: generating a texture analysis image set according to the abnormal area.

[0258] It is understood that all processed images are collated and aggregated to generate a complete texture analysis image set, which shows the possible defect locations and features on the surface of photovoltaic modules.

[0259] Step S44, obtaining a fault image set according to at least one of the edge detection image, the hot spot detection image set and the texture analysis image set.

[0260] As you can see, edge detection images help identify linear defects on PV modules, such as small cracks and scratches, where the edges of these defects are clearly captured and marked in the image. Hot spot detection images can directly point out the hot areas of the module, which may have electrical faults or material damage. Detecting these areas helps prevent further damage to the module in advance. Texture analysis images can identify subtle texture changes caused by aging, scratches, material defects, etc., especially those areas that are difficult to detect with the naked eye. All processed images are organized into a fault image set to provide an intuitive and comprehensive fault reference for subsequent maintenance and inspection.

[0261] This embodiment provides a multi-camera fusion inspection method for photovoltaic stations. By combining Canny edge detection, infrared thermal imaging analysis and local binary pattern (LBP) texture feature extraction, it solves the technical problem of difficulty in detecting tiny cracks, hot spots and other surface defects in photovoltaic stations. Through these technical means, this embodiment can accurately identify the edge features, temperature abnormality areas and texture changes of photovoltaic modules, and generate a set of fault images containing these features. This not only improves the accuracy and reliability of photovoltaic module fault detection, but also speeds up the speed of fault identification, and provides intuitive and comprehensive fault references for photovoltaic station operation and maintenance personnel, thereby realizing effective monitoring of the health status of photovoltaic stations, ensuring the stable operation of photovoltaic power generation systems and extending the service life of equipment.

[0262] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the multi-camera fusion inspection method for photovoltaic stations of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0263] This application also provides a photovoltaic station multi-camera fusion inspection device, please refer to Figure 6 The photovoltaic station multi-camera fusion inspection device comprises:

[0264] An acquisition module 10 is used to acquire an initial image set of a photovoltaic station field captured by multiple cameras;

[0265] A calibration module 20, configured to calibrate the timestamp and spatial viewing angle of the initial image set to obtain a calibrated image set;

[0266] A preprocessing module 30, configured to perform image preprocessing according to the calibration image set to obtain an enhanced image set;

[0267] The analysis module 40 is used to perform feature extraction analysis based on the enhanced image set to obtain a fault image set, and complete the inspection of the photovoltaic station based on the fault image set.

[0268] The photovoltaic station multi-camera fusion inspection device provided by the present application adopts the photovoltaic station multi-camera fusion inspection method in the above-mentioned embodiment, which can solve the technical problem that the image quality collected by the camera is affected by environmental factors such as lighting changes, reflections and glare in photovoltaic component monitoring, and there is a risk of missed detection and false detection. Compared with the prior art, the beneficial effects of the photovoltaic station multi-camera fusion inspection device provided by the present application are the same as the beneficial effects of the photovoltaic station multi-camera fusion inspection method provided by the above-mentioned embodiment, and the other technical features of the photovoltaic station multi-camera fusion inspection device are the same as the features disclosed in the above-mentioned embodiment method, which will not be repeated here.

[0269] In one embodiment, the analysis module 40 is further used to perform edge detection on the enhanced image set to obtain an edge detection image set; convert the pixel values ​​of the enhanced image set into a target temperature, and count the size of the hot spot area according to a preset temperature and the target temperature, and obtain a hot spot detection image set according to the statistical results; extract texture features of the enhanced image set according to a local binary pattern, and obtain a texture analysis image set according to the texture features; and obtain a fault image set according to at least one of the edge detection image, the hot spot detection image set, and the texture analysis image set.

[0270] In one embodiment, the analysis module 40 is also used to calculate the gradient of the image in the texture analysis image set to obtain the gradient amplitude and gradient direction of the image; obtain a first threshold and a second threshold, and obtain an edge candidate area based on the first threshold, the second threshold, the gradient amplitude and the gradient direction; perform non-maximum suppression on the edge candidate area to obtain an initial edge image; perform edge connection and filtering on the initial edge image to obtain an edge detection image set.

[0271] In one embodiment, the analysis module 40 is further used to obtain the central pixel of each image in the enhanced image set, and determine the number of pixels and the neighborhood radius of the local binary pattern according to the central pixel; compare the gray value of each pixel with the gray value of its surrounding neighborhood pixels according to the number of pixels and the neighborhood radius to obtain a comparison result; generate a binary texture value according to the comparison result, and obtain a texture feature map according to the binary texture value; detect abnormal areas according to texture changes in the texture feature map;

[0272] A texture analysis image set is generated according to the abnormal area.

[0273] In one embodiment, the preprocessing module 30 is further used to perform denoising on the calibration image set according to Gaussian filtering to obtain an initial denoised image set; when a faulty image is detected in the initial denoised image set, perform denoising on the faulty image according to bilateral filtering to obtain a target denoised image set; perform adaptive contrast enhancement on the target denoised image set according to adaptive histogram equalization to obtain a contrast enhanced image set; analyze the brightness distribution of the contrast enhanced image set to obtain an analysis result, and compress the highlight area in the contrast enhanced image set or enhance the brightness of the dark area in the contrast enhanced image set according to the analysis result to obtain an enhanced image set.

[0274] In one embodiment, the preprocessing module 30 is further used to divide each image in the target denoised image set into sub-blocks of a preset size according to adaptive histogram equalization; count the number of pixels in the sub-blocks to obtain a histogram, calculate the pixel cumulative value of the histogram, perform equalization according to the pixel cumulative value to obtain an equalized pixel value; adjust the equalized pixel value according to a contrast limiting mechanism to obtain an enhanced local area; and merge the enhanced local areas to obtain a contrast enhanced image set.

[0275] In one embodiment, the calibration module 20 is further used to obtain image timestamps and camera codes of images in the initial image set, and sort the image timestamps according to the camera codes to obtain sorted timestamps; calibrate the image timestamps according to the sorted timestamps and the camera codes to obtain a time-series synchronized image set; construct integral images for the images in the time-series synchronized image set, and detect feature points in the integral images according to the Hessian matrix; calculate the distances between feature points in different integral images, and match the feature points of the different integral images according to the distances to obtain initial matching points; filter invalid matching points in the initial matching points according to a ratio test to obtain target matching points; calculate a correction matrix according to the target matching points, and normalize according to the correction matrix to obtain a calibration image set.

[0276] The present application provides a multi-camera fusion inspection device for a photovoltaic station, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the multi-camera fusion inspection method for the photovoltaic station in the above-mentioned embodiment one.

[0277] Reference below Figure 7 , which shows a schematic diagram of the structure of a photovoltaic station multi-camera fusion inspection device suitable for implementing the embodiment of the present application. The photovoltaic station multi-camera fusion inspection device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The photovoltaic station multi-camera fusion inspection equipment shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0278] like Figure 7As shown, the photovoltaic station multi-camera fusion inspection equipment may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 to the random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the photovoltaic station multi-camera fusion inspection equipment are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the photovoltaic field station multi-camera fusion inspection device to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a photovoltaic field station multi-camera fusion inspection device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.

[0279] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0280] The photovoltaic station multi-camera fusion inspection equipment provided by the present application adopts the photovoltaic station multi-camera fusion inspection method in the above embodiment, which can solve the technical problem that the image quality collected by the camera is affected by environmental factors such as lighting changes, reflections and glare in photovoltaic component monitoring, and there is a risk of missed detection and false detection. Compared with the prior art, the beneficial effects of the photovoltaic station multi-camera fusion inspection equipment provided by the present application are the same as the beneficial effects of the photovoltaic station multi-camera fusion inspection method provided by the above embodiment, and the other technical features of the photovoltaic station multi-camera fusion inspection equipment are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0281] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0282] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0283] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the photovoltaic station multi-camera fusion inspection method in the above-mentioned embodiment.

[0284] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0285] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned photovoltaic station multi-camera fusion inspection method, and can solve the technical problem that the image quality collected by the camera is affected by environmental factors such as lighting changes, reflections and glare in photovoltaic module monitoring, and there is a risk of missed detection and false detection. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the photovoltaic station multi-camera fusion inspection method provided by the above-mentioned embodiment, and will not be repeated here.

[0286] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned photovoltaic station multi-camera fusion inspection method.

[0287] The computer program product provided by the present application can solve the technical problem that the image quality collected by the camera is affected by environmental factors such as lighting changes, reflections and glare, and there is a risk of missed detection and false detection in photovoltaic module monitoring. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the photovoltaic station multi-camera fusion inspection method provided by the above embodiment, which will not be repeated here.

[0288] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A photovoltaic station multi-camera fusion inspection method, characterized in that: The method comprises: Obtaining an initial set of images of the photovoltaic station field taken by multiple cameras; Calibrate the timestamp and spatial perspective of the initial image set to obtain a calibrated image set; Performing image preprocessing according to the calibration image set to obtain an enhanced image set; Feature extraction and analysis are performed based on the enhanced image set to obtain a fault image set, and the inspection of the photovoltaic station is completed based on the fault image set.

2. The method according to claim 1, characterized in that The step of performing feature extraction and analysis according to the enhanced image set to obtain a fault image set comprises: Performing edge detection on the enhanced image set to obtain an edge detection image set; Converting the pixel values ​​of the enhanced image set into a target temperature, and counting the size of the hot spot area according to a preset temperature and the target temperature, and obtaining a hot spot detection image set according to the statistical results; Extracting texture features of the enhanced image set according to local binary patterns, and obtaining a texture analysis image set according to the texture features; A fault image set is obtained according to at least one of the edge detection image, the hot spot detection image set and the texture analysis image set.

3. The method according to claim 2, characterized in that The step of performing edge detection on the texture analysis image set to obtain an edge detection image set comprises: Calculating the gradient of an image in the texture analysis image set to obtain a gradient magnitude and a gradient direction of the image; Obtaining a first threshold and a second threshold, and obtaining an edge candidate region according to the first threshold, the second threshold, the gradient amplitude, and the gradient direction; Performing non-maximum suppression on the edge candidate region to obtain an initial edge image; Edge connection and filtering are performed on the initial edge image to obtain an edge detection image set.

4. The method according to claim 2, characterized in that The step of extracting texture features of the enhanced image set according to the local binary pattern and obtaining the texture analysis image set according to the texture features comprises: Obtaining a central pixel of each image in the enhanced image set, and determining the number of pixels and the neighborhood radius of the local binary pattern according to the central pixel; According to the number of pixels and the neighborhood radius, the grayscale value of each pixel is compared with the grayscale values ​​of its surrounding neighborhood pixels to obtain a comparison result; Generate a binary texture value according to the comparison result, and obtain a texture feature map according to the binary texture value; Detect abnormal areas based on texture changes in texture feature maps; A texture analysis image set is generated according to the abnormal area.

5. The method according to claim 1, characterized in that The image preprocessing includes noise reduction, adaptive contrast enhancement and brightness adjustment; the step of performing image preprocessing according to the calibration image set to obtain an enhanced image set includes: Performing denoising on the calibration image set according to Gaussian filtering to obtain an initial denoised image set; When it is detected that there is a fault image in the initial denoised image set, denoising the fault image according to bilateral filtering to obtain a target denoised image set; Adaptively contrast enhancing the target denoised image set according to adaptive histogram equalization to obtain a contrast enhanced image set; The brightness distribution of the contrast enhanced image set is analyzed to obtain an analysis result, and according to the analysis result, the highlight area in the contrast enhanced image set is compressed or the brightness of the dark area in the contrast enhanced image set is enhanced to obtain an enhanced image set.

6. The method according to claim 5, characterized in that The step of performing adaptive contrast enhancement on the target denoised image set according to adaptive histogram equalization to obtain a contrast enhanced image set comprises: Dividing each image in the target denoised image set into sub-blocks of a preset size according to adaptive histogram equalization; Counting the number of pixels of the sub-block to obtain a histogram, calculating a pixel cumulative value of the histogram, and performing equalization according to the pixel cumulative value to obtain an equalized pixel value; Adjusting the equalized pixel value according to a contrast limiting mechanism to obtain an enhanced local area; The enhanced local regions are merged to obtain a contrast enhanced image set.

7. The method according to claim 1, characterized in that The step of calibrating the timestamp and spatial perspective of the initial image set to obtain a calibrated image set comprises: Obtaining image timestamps and camera codes of images in the initial image set, and sorting the image timestamps according to the camera codes to obtain sorted timestamps; Calibrate the image timestamp according to the sorting timestamp and the camera code to obtain a time-series synchronized image set; constructing an integral image for the images in the time-series synchronous image set, and detecting feature points in the integral image according to a Hessian matrix; Calculating distances between feature points in different integral images, and matching the feature points of the different integral images according to the distances to obtain initial matching points; Filter invalid matching points from the initial matching points according to a ratio test to obtain target matching points; A correction matrix is ​​calculated according to the target matching points, and normalization is performed according to the correction matrix to obtain a calibration image set.

8. A photovoltaic station multi-camera fusion inspection device, characterized in that: The device comprises: An acquisition module is used to acquire an initial image set of a photovoltaic station field captured by multiple cameras; A calibration module, used to calibrate the timestamp and spatial viewing angle of the initial image set to obtain a calibrated image set; A preprocessing module, used for performing image preprocessing according to the calibration image set to obtain an enhanced image set; The analysis module is used to perform feature extraction analysis based on the enhanced image set to obtain a fault image set, and complete the inspection of the photovoltaic station based on the fault image set.

9. A photovoltaic station multi-camera fusion inspection device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the photovoltaic station multi-camera fusion inspection method according to any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the photovoltaic station multi-camera fusion inspection method according to any one of claims 1 to 7 are implemented.

Images