Image processing method and photovoltaic detection robot

Abstract

An image processing method and a photovoltaic detection robot are disclosed. The method is used for optimizing an initial image of a to-be-detected photovoltaic module, and comprises the following steps: smoothing the initial image to obtain a first image of the to-be-detected photovoltaic module, the first image comprising low-frequency information of the initial image; performing high-pass filtering on the initial image based on the first image to obtain a second image of the photovoltaic module to be detected, the second image including high-frequency information of the initial image; the first image and the second image are split into a plurality of sub-image blocks, histogram equalization operation is carried out on each sub-image block based on a contrast increase threshold, the first image comprises a first sub-image block, and the second image comprises a second sub-image block; the contrast increment of the second sub-image block is greater than that of the first sub-image block; and synthesizing the processed image so as to highlight the detail part of the photovoltaic module to be detected. Through the method, the image of the photovoltaic module can be optimized, and the defect detection efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and particularly to an image processing method and a photovoltaic detection robot. Background Art

[0002] The detection of photovoltaic modules is crucial because, as the core components of a solar power generation system, the quality of photovoltaic modules directly affects the power generation efficiency and system lifespan. In practical applications, photovoltaic modules may be affected by mechanical stress or the external environment during manufacturing, transportation, and installation, resulting in problems such as hidden cracks, short circuits, and welding defects. These defects not only reduce the conversion efficiency of photovoltaic modules but may also cause local overheating (hot spot effect), further accelerating the aging of the modules and even posing safety hazards. Therefore, timely detection and repair of these problems are crucial for improving system performance and extending the service life.

[0003] Among existing detection methods, the EL (electroluminescence) detection technology is an efficient and accurate method that can clearly display defects such as hidden cracks and dark spots inside photovoltaic modules. However, EL detection also faces certain limitations, such as insufficient ability to identify fine cracks and minor defects. Especially in an image environment with low contrast and high noise, these defects are easily masked or misjudged, posing challenges to component quality control.

[0004] Therefore, how to optimize the image of a photovoltaic module to be detected during electroluminescence detection and improve the efficiency of defect detection is an urgent problem to be solved. Summary of the Invention

[0005] In view of this, this application discloses an image processing method and a photovoltaic detection robot to optimize the image of a photovoltaic module to be detected during electroluminescence detection, thereby improving the efficiency of defect detection.

[0006] In a first aspect, the present application discloses an image processing method for optimizing an initial image of a photovoltaic module to be detected. The initial image is an image of the photovoltaic module to be detected during electroluminescence detection. The method includes: performing smoothing processing on the initial image to obtain a first image of the photovoltaic module to be detected, where the first image includes the low-frequency information of the initial image; performing high-pass filtering on the initial image based on the first image to obtain a second image of the photovoltaic module to be detected, where the second image includes the high-frequency information of the initial image; respectively splitting the first image and the second image into a plurality of sub-image blocks, and performing histogram equalization operations on each sub-image block based on a contrast increase threshold. The first image includes first sub-image blocks, the second image includes second sub-image blocks, and the contrast increase amount of the second sub-image blocks is greater than that of the first sub-image blocks; determining a weight value according to the information of the initial image, and synthesizing a third image and a fourth image based on the weight value to highlight the detailed parts of the photovoltaic module to be detected. The third image is the image obtained after the histogram equalization operation on the first image, and the fourth image is the image obtained after the histogram equalization operation on the second image.

[0007] Optionally, the initial image includes a first pixel; performing smoothing processing on the initial image to obtain a first image of the photovoltaic module to be detected includes: performing Gaussian filtering on the initial image; determining a first pixel domain based on a preset radius with the first pixel as the center; performing Gaussian blur processing on the first pixel domain using a convolution kernel with a preset standard deviation to obtain the smoothed first image.

[0008] Optionally, the image processing method further includes: setting a plurality of candidate standard deviations, and respectively determining the performance of each candidate standard deviation during Gaussian blur processing on the first pixel domain; determining a preset standard deviation from the plurality of candidate standard deviations based on a first preset index; where the first preset index includes at least one of the performance indexes during Gaussian blur processing on the first pixel domain: signal-to-noise ratio, structural similarity of the photovoltaic module to be detected, edge retention of the photovoltaic module to be detected, or mean square error.

[0009] Optionally, the initial image further includes a second pixel, and the first image includes a third pixel. Performing high-pass filtering on the initial image based on the first image to obtain a second image of the photovoltaic module to be detected includes: determining the pixel value of a fourth pixel by subtracting the pixel value of the third pixel from the pixel value of the second pixel, and the second image includes the fourth pixel.

[0010] Optionally, the first image includes a first number of first sub-image blocks, and the second image includes a second number of second sub-image blocks, where the first number is less than the second number; splitting the first image and the second image into a number of sub-image blocks respectively, including: determining the first number and the second number based on the image resolution and the size of the target area, where the image resolution includes the image resolution of the first image and the image resolution of the second image, and the size of the target area is the size of the defect on the photovoltaic module to be detected; splitting the first image into the first number of first sub-image blocks based on the first number, where the sizes of all the first sub-image blocks are the same; splitting the second image into the second number of second sub-image blocks based on the second number, where the sizes of all the second sub-image blocks are the same.

[0011] Optionally, performing histogram equalization operations on each sub-image block based on a contrast increase threshold, including: calculating the histogram of each sub-image block, and processing the first sub-image block based on the contrast increase threshold; calculating the cumulative distribution function of each processed sub-image block; remapping the pixel values in each sub-image block using the cumulative distribution function based on the maximum gray level number and the maximum and minimum values of the cumulative distribution function, so as to enhance the local contrast of the low-frequency information of the first image and the high-frequency information of the second image.

[0012] Optionally, the image processing method further includes: setting a number of candidate thresholds, and respectively determining the performance of each candidate threshold when performing histogram equalization processing on the sub-image block; determining the contrast increase threshold from the number of candidate thresholds based on a second preset index; where the second preset index includes at least one of the performance indexes for performing histogram equalization processing on the sub-image block: signal-to-noise ratio, image contrast, local contrast mean, edge intensity or edge sharpness.

[0013] Optionally, the image processing method further includes: reducing the first number to reduce the contrast of each first sub-image block, so that the low-frequency information of the first image is uniformly reflected; increasing the second number to enhance the contrast of each second sub-image block, so that the high-frequency information of the second image is prominently reflected.

[0014] Optionally, determining a weight value according to the information of the image, and synthesizing the third image and the fourth image based on the weight value, so that the detailed part of the photovoltaic module to be detected is prominent, including: determining a first area and a second area based on the information of the initial image; increasing the weight value of the first image in the first area and increasing the weight value of the second image in the second area, so as to prominently display the border and the detailed distribution of the photovoltaic module to be detected.

[0015] In a second aspect, the present application discloses a photovoltaic detection robot, including: an image acquisition module configured to acquire an initial image of a photovoltaic module to be detected; and a processor configured to receive the initial image and apply the image processing method disclosed in the above first aspect to optimize the initial image.

[0016] In summary, the image processing method and the photovoltaic detection robot disclosed in the present application have at least the following

[0017] Advantages:

[0018] (1) By separately extracting the high-frequency information and the low-frequency information in the initial image and then adopting different strategies for enhancement processing, not only the regional information such as the photovoltaic module area and the border is effectively retained, but also the visibility of the details in the image of the photovoltaic module to be detected is significantly improved, especially the tiny cracks and other defects that are difficult to identify, thereby enhancing the accuracy of defect detection.

[0019] (2) Gaussian blur processing is used as a preprocessing step to reduce the noise in the image to be detected, thereby reducing the interference of noise on defect recognition in subsequent processing.

[0020] (3) By enhancing the contrast of local regions of the image, the problem of over-enhancement that may be caused by traditional global histogram equalization is solved, enabling the local details of the image to be clearly displayed and the image to be distortion-free.

[0021] (4) The present application is not only applicable to the EL detection of new photovoltaic panels but also can be used for the maintenance detection of old photovoltaic panels, and has wide applicability in the photovoltaic industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The following briefly introduces the drawings used in the description of the embodiments of the present application:

[0023] Figure 1 is a flowchart of an image processing method provided by an embodiment of the present application.

[0024] Figure 2 is a structural schematic diagram of a photovoltaic detection robot provided by an embodiment of the present application.

[0025] In the figure: photovoltaic detection robot - 200, image acquisition module - 210, processor - 220, photovoltaic module to be detected - 230. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] To more clearly illustrate the technical solutions of the embodiments of the present application, the specific implementation manners of the present application will be described below with reference to the accompanying drawings. The accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings, and other implementation manners can be obtained. Adjustments and improvements made without departing from the concept of the present application all fall within the protection scope of the present application.

[0027] To make the drawings concise, only the parts related to the corresponding embodiments are schematically shown in each drawing, and they do not represent the actual structure of the product. In addition, to make the drawings concise and easy to understand, in some drawings, only some of the components with the same structure or function are schematically shown, and there may actually be more or fewer components with the same structure or function.

[0028] In the present application, unless otherwise clearly specified and limited, ordinal numbers, such as "first", "second", etc., are only used to distinguish and describe related objects, and cannot be understood as indicating or implying the relative importance or order between related objects; in addition, they do not represent the quantity of related objects. "Multiple" includes two or more, and other quantifiers are similar. " / " is used to describe the relationship between related objects, which means the "or" relationship between related objects. "And / or" is used to describe the relationship between related objects, which includes any combination relationship between related objects. For example, "a and / or b" includes: "a alone", "b alone", or "a and b". "One or more" or "at least one" among multiple objects refers to any object or any combination of multiple objects. For example, "one or more of a1, a2, a3" or "at least one of a1, a2, a3" includes: "a1 alone", "a2 alone", "a3 alone", "a1 and a2", "a1 and a3", "a2 and a3", or "a1, a2 and a3".

[0029] The detection of photovoltaic modules (such as photovoltaic panels) is crucial for the stable operation and efficient power generation of solar power systems. As an important part of clean energy, photovoltaic power generation systems have been widely used globally. However, due to the relatively harsh working environment of photovoltaic panels, such as in desert, plateau, coastal and other regions, their surfaces are extremely vulnerable to damage by external environmental factors. Conditions such as high temperature, strong wind and sand, salt mist, and large day-night temperature differences in these environments will accelerate the surface wear and internal material aging of photovoltaic panels. After long-term exposure to such environments, cracks, scratches, and even material peeling may appear on the surface of photovoltaic panels. In addition, during the manufacturing, transportation, and installation of photovoltaic modules, mechanical stress or improper operation can also cause defects such as hidden cracks, poor soldering, and damaged encapsulation layers. If these defects are not detected and repaired in time, they will not only lead to a decrease in the power generation efficiency of photovoltaic panels, but may also cause local overheating (hot spot effect), further damaging the component lifespan and even increasing the fire hazard. Therefore, regular detection and maintenance of photovoltaic modules are important ways to improve their reliability.

[0030] In existing photovoltaic module detection technologies, EL detection is widely used in the quality control and maintenance detection of the photovoltaic industry. EL detection applies a forward voltage or current to the photovoltaic module to make the internal solar cells emit electroluminescence signals, and uses a high-sensitivity imaging device to capture these signals to generate images. These images can intuitively reflect hidden cracks, solder joint defects, dark spots, and other potential problems inside the photovoltaic module. Compared with traditional visual inspection, EL detection has higher resolution and can detect subtle defects that are difficult to detect by the naked eye. Therefore, it is considered one of the standard methods for photovoltaic module defect detection.

[0031] However, although EL detection plays an important role in the production and maintenance of photovoltaic modules, there are still some limitations in its practical applications. First, the ability of EL detection to identify subtle cracks and other minor defects is relatively weak. Such defects usually appear as faint dark lines or small-scale brightness variations in the image, which are not obvious enough in the image and are easily masked by noise. In addition, there may be external interferences such as stains, scratches, or optical reflections on the surface of the photovoltaic module, which will further reduce the contrast of the image and increase the difficulty of defect identification. Especially in an image environment with low contrast and high noise, existing EL detection algorithms are difficult to effectively extract the features of these tiny defects, resulting in an increased probability of false detection or missed detection. Second, the workflow of traditional EL detection relies heavily on manual analysis. Although this method can utilize experience and subjective judgment to discover defects in some cases, it is inefficient and difficult to meet the requirements of large-scale photovoltaic module detection. With the rapid development of the photovoltaic industry, the production and installation scale of modules have been increasing year by year, and it has become difficult to cope with the high-efficiency detection requirements of modernization relying solely on manual analysis. In addition, the subjectivity of manual analysis is relatively strong, and different technicians may have inconsistent judgment results for the same defect, further increasing the uncertainty of the detection results.

[0032] Based on the deficiencies existing in the prior art, the concept of this application lies in achieving an efficient image enhancement strategy, including the separation and optimization processing of high-frequency information and low-frequency information, which significantly improves the visibility of details such as hidden cracks in the image, thereby improving the accuracy and reliability of defect detection. That is to say, by optimizing the image of the photovoltaic module to be detected during electroluminescence detection, the efficiency of defect detection is further improved.

[0033] The following is a description with reference to the accompanying drawings.

[0034] Figure 1 It is a flowchart of an image processing method provided by an embodiment of this application. Please refer to Figure 1 An image processing method for optimizing the initial image of a photovoltaic module to be detected, where the initial image is the image of the photovoltaic module to be detected during electroluminescence detection. The method includes:

[0035] S100, performing smoothing processing on the initial image to obtain the first image of the photovoltaic module to be detected, where the first image includes the low-frequency information of the initial image;

[0036] S200, performing high-pass filtering on the initial image based on the first image to obtain the second image of the photovoltaic module to be detected, where the second image includes the high-frequency information of the initial image;

[0037] In S300, the first image and the second image are respectively split into a number of sub-image blocks, and histogram equalization operations are performed on each sub-image block based on a contrast increase threshold. The first image includes a first sub-image block, and the second image includes a second sub-image block. The contrast increase amount of the second sub-image block is greater than that of the first sub-image block.

[0038] In S400, weight values are determined according to the information of the initial image, and the third image and the fourth image are synthesized based on the weight values to highlight the detailed parts of the photovoltaic module to be detected. The third image is the image obtained after the first image completes the histogram equalization operation, and the fourth image is the image obtained after the second image completes the histogram equalization operation.

[0039] The target image to be processed in this application is the image of the photovoltaic module to be detected during electroluminescence detection. As mentioned above, by applying a forward voltage or current to the photovoltaic module, the solar cells inside emit electroluminescence signals, and a high-sensitivity imaging device is used to capture these signals to generate images; for the sake of convenience of description, these images are called initial images. The initial images include both high-frequency information and low-frequency information.

[0040] Among them, low-frequency information has the following common characteristics in image processing and analysis: low-frequency information usually covers a relatively large spatial area in the image, and compared with specific details, it describes the overall shape and structure more; within the area of low-frequency information, the changes in color and brightness are relatively gentle, without sudden jumps; low-frequency information is very important for identifying the main objects and components in the image, and it helps to establish the basic framework and background of the scene. Exemplarily, the contours and large structural areas in the initial image are low-frequency information.

[0041] Correspondingly, high-frequency information refers to the detailed parts in the image, such as edges, lines, textures, and patterns. These features usually appear in areas where the pixel values change rapidly, such as the boundaries of objects or surfaces with complex textures. In the areas of cracks, scratches, or dust-covered areas on the photovoltaic panel, high-frequency information is particularly obvious because the pixel values change significantly within a very small range. Compared with low-frequency information, high-frequency information usually appears as the part where the pixel values in the image change rapidly, and its contrast is relatively high, making these areas more prominent visually and helping to identify and analyze the specific features of the object. Since high-frequency information contains the detailed and textured parts of the image, they are particularly sensitive to noise. During image acquisition and processing, noise usually affects these high-frequency components first, so it is necessary to suppress noise and reduce the loss of details simultaneously during preprocessing. Although the spatial proportion of high-frequency information in the image is small, its role in vision is very important. In image recognition, classification, and analysis, high-frequency information can provide key visual clues, such as when identifying small damages (such as hidden cracks) on the photovoltaic panel.

[0042] Since the low-frequency information and high-frequency information represent different meanings respectively, after separating the low-frequency information and high-frequency information of the initial image and processing them separately, different features of the image can be enhanced more pertinently, thereby improving the overall processing effect. Exemplarily, since noise mainly exists in the high-frequency part, directly processing the initial image may cause the noise to be amplified; after separating the high-frequency information, these areas can be selectively processed, for example, using a denoising algorithm to suppress high-frequency noise and then enhancing the real details; while the noise in the low-frequency image is relatively less, more attention can be paid to improving the brightness uniformity and structural integrity.

[0043] The low-frequency information and high-frequency information of the initial image can be obtained by using smoothing processing and high-pass filtering respectively. The purpose of smoothing processing is to eliminate the high-frequency components (such as edges, details, noise) in the image and retain the low-frequency components (such as the overall shape and large-scale brightness distribution) with slower changes in the image. The purpose of high-pass filtering is to highlight the high-frequency components (such as edges, details, textures) in the image and suppress the low-frequency components (such as the background and uniform areas). For the convenience of description, the image obtained after smoothing the initial image is called the first image (or the low-frequency image), and the first image includes the low-frequency information of the initial image; the image obtained after high-pass filtering is called the second image (or the high-frequency image), and the second image includes the high-frequency information of the initial image. The resolutions of the first image, the second image, and the initial image are the same, that is, the number of pixels of the three is the same.

[0044] Subsequently, the first image and the second image are respectively split into a number of sub-image blocks. For example, the first image is split into a first number of first sub-image blocks, and the second image is split into a second number of second sub-image blocks. The first number and the second number can be the same or different. When the first number and the second number are the same, it means that the areas covered by each sub-image block in the high-frequency image and the low-frequency image exactly correspond, which enables direct paired analysis of the high-frequency information and low-frequency information in the same area during subsequent processing and simplifies the calculation logic. When the first number and the second number are different, for example, the first number is less than the second number. At this time, since the resolutions of the first image and the second image are the same, the area of a first sub-image block is larger than the area of a second sub-image block. That is to say, by splitting the high-frequency image into more small blocks, finer enhancement can be performed on the details; by splitting the low-frequency image into fewer large blocks, more attention can be paid to the smoothing and uniformity of the overall area. After splitting, histogram equalization operations are performed on each sub-image block based on a contrast increase threshold, and the contrast increase amount of the second sub-image block is greater than that of the first sub-image block. In this way, the details and textures of the image can be further highlighted to facilitate the display of details such as hidden cracks.

[0045] After separately enhancing the contrast of the high-frequency image and the low-frequency image, the two are fused with certain weights respectively. The obtained image after fusion is the optimized image of the original image. For ease of description, it is referred to as the final image. Compared with the original image, the final image increases the visibility of details while maintaining the overall stability of the image. The weight values used for fusion can be determined from the information of the original image. In some embodiments of the present application, the weight value of the first image is less than the weight value of the second image. That is to say, the weight value of the low-frequency image is less than the weight value of the high-frequency image. Exemplarily, the weight value of the low-frequency image is 0.3, and the weight value of the high-frequency image is 0.7. Since visually, high-frequency information is more important than low-frequency information, assigning a larger weight value to the high-frequency image can better optimize the original image, thereby improving the detection efficiency.

[0046] In some embodiments of the present application, the weight values are determined according to the information of the original image, and the third image and the fourth image are synthesized based on the weight values to highlight the detailed parts of the photovoltaic module to be detected, including: determining a first region and a second region based on the information of the original image; increasing the weight value of the first image in the first region and increasing the weight value of the second image in the second region to highlight the border and detailed distribution of the photovoltaic module to be detected. For example, the first region may be a region with rich details. For this region, the weight of high-frequency information can be increased; another example is that the second region may be a relatively uniform region. For this region, the weight of low-frequency information can be increased. Or, a more refined region-aware processing strategy can be implemented, applying different enhancement strategies to different parts of the image. For example, more aggressive high-frequency enhancement is used for the damaged region, while more low-frequency smoothing is used for the background region. Among them, the information of the original image may include the detailed distribution information of the photovoltaic module to be detected or the damage distribution information of the photovoltaic module to be detected.

[0047] In some embodiments of the present application, deep learning techniques can be combined to preprocess and enhance the image. Exemplarily, a trained neural network is used to identify and enhance the key features of the photovoltaic module, such as cracks, stains, etc. In this way, high-frequency and low-frequency information can be distinguished more accurately, and the processing effect can be optimized by learning a large amount of data.

[0048] In some embodiments of the present application, the original image includes a first pixel; smoothing the original image to obtain a first image of the photovoltaic module to be detected, including: performing Gaussian filtering on the original image; determining a first pixel domain based on a preset radius centered on the first pixel; performing Gaussian blur processing on the first pixel domain using a convolution kernel with a preset standard deviation to obtain the smoothed first image.

[0049] Gaussian blur is achieved by weighted summation of the convolution kernel and the pixels around each pixel of the image. The calculation formula is as follows: where G(x, y) is the Gaussian function centered at the point (x, y); σ is the standard deviation, which is used to control the degree of blur. The updated value of each pixel point in the image is the weighted sum of the surrounding pixel values and the weights of the Gaussian function. That is to say, the initial image includes the first pixel, and the point (x, y) is the coordinate point of the first pixel. In units of pixels, by performing Gaussian blur processing on each pixel in the initial image, the smoothing of the entire initial image is completed, thereby separating the low-frequency information in the initial image, that is, the first image. Among them, the first pixel domain is jointly determined by the first pixel and the preset radius, and the preset radius is determined by the size of the convolution kernel. Before performing Gaussian blur processing, Gaussian filtering is performed on the initial image, which can eliminate high-frequency noise and improve the quality of low-frequency information extraction, that is, by smoothing out random noise and high-frequency textures, the low-frequency extraction basis is optimized. Retaining low-frequency information means that the boundaries and overall layout of the plates can be clearly seen, which is crucial for subsequent damage location and identification.

[0050] In some embodiments of the present application, the image processing method further includes: setting a plurality of candidate standard deviations, and respectively determining the performance of each candidate standard deviation when performing Gaussian blur processing on the first pixel domain; determining a preset standard deviation among the plurality of candidate standard deviations based on a first preset index; wherein, the first preset index includes at least one of the performance indexes of Gaussian blur processing on the first pixel domain: signal-to-noise ratio, structural similarity of the photovoltaic module to be detected, edge retention degree of the photovoltaic module to be detected, or mean square error.

[0051] As one of the key parameters of Gaussian blur processing, the standard deviation determines the degree of blur of the image. The larger the standard deviation, the higher the degree of blur. The purpose of Gaussian blur is to smooth the image, reduce noise and details. The size of the convolution kernel determines the intensity and effect of smoothing: when the convolution kernel is small, the smoothing degree is low and more details are retained; when the convolution kernel is large, more distant pixels are involved, and the smoothing effect is stronger, but details may be lost. The size of the convolution kernel is generally used in conjunction with the standard deviation of the Gaussian function. The larger the standard deviation, the larger the convolution kernel is required to cover a sufficient pixel range.

[0052] Before determining the preset standard deviation, several candidate standard deviations can be set first. Their values can be, for example, 1, 3, 5, etc. Initially, a relatively small standard deviation can be set, such as 1, to observe its effect, and then gradually increased, such as 3 or 5, etc. Each time, observe the impact of Gaussian blur on the image until a balance point is found, which provides the best effect between reducing noise and retaining low-frequency information. That is to say, the most suitable candidate standard deviation can be selected through certain preset metrics, and this is used as the preset standard deviation for Gaussian blur processing. For the sake of distinction, the preset metric here is called the first preset metric, and the first preset metric includes at least one of the performance metrics in Gaussian blur processing of the first pixel domain: signal-to-noise ratio, structural similarity of the photovoltaic module to be detected, edge retention of the photovoltaic module to be detected, or mean square error. For example, calculate the signal-to-noise ratio of the image after Gaussian blur and select the standard deviation value with the highest signal-to-noise ratio; or, compare the structural similarity between the first image after Gaussian blur and the initial image to find the standard deviation that can both reduce noise and retain contour information; or, use an edge detection algorithm to extract edges, evaluate the edge strength after blur, and select the standard deviation with the least loss of edge information; or, calculate the mean square error between the first image after blur and the initial image to optimize the balance between noise suppression and detail retention. Through the above metrics, the optimal standard deviation can be quantitatively obtained and used as the preset standard deviation.

[0053] In some embodiments of the present application, the initial image further includes a second pixel, and the first image includes a third pixel. High-pass filtering the initial image based on the first image to obtain a second image of the photovoltaic module to be detected, including: determining the pixel value of a fourth pixel by subtracting the pixel value of the third pixel from the pixel value of the second pixel, and the second image includes the fourth pixel.

[0054] The initial image includes a second pixel, and the first image includes a third pixel. The relative position of the second pixel in the initial image matches the relative position of the third pixel in the first image. For example, the relative position of the second pixel in the initial image is (2, 3), and the relative position of the third pixel in the first image is also (2, 3), and the two match each other. The second image includes the fourth pixel, and the relative position of the fourth pixel in the second image matches the relative position of the third pixel in the first image. For example, the relative position of the fourth pixel in the second image is also (2, 3). In this way, by subtracting the pixel value of the third pixel from the pixel value of the second pixel, the pixel value of the fourth pixel is obtained. In this way, in units of pixels, the pixel values of each pixel in the second image can be obtained, and thus the information of the entire second image is determined. By calculating the difference between the initial image and the first image after Gaussian blur, high-frequency information (i.e., details and textures) is extracted and used as the second image, so that the second image includes the high-frequency information of the initial image.

[0055] In some embodiments of the present application, the first image includes a first number of first sub-image blocks, the second image includes a second number of second sub-image blocks, and the first number is less than the second number; splitting the first image and the second image into a number of sub-image blocks respectively, including: determining the first number and the second number based on the image resolution and the size of the target area, the image resolution including the image resolution of the first image and the image resolution of the second image, and the size of the target area being the size of the defect on the photovoltaic module to be detected; splitting the first image into the first number of first sub-image blocks based on the first number, and the sizes of all the first sub-image blocks are the same; splitting the second image into the second number of second sub-image blocks based on the second number, and the sizes of all the second sub-image blocks are the same.

[0056] The algorithm for enhancing the contrast of the first image and the second image can be the contrast limited adaptive histogram equalization (CLAHE) algorithm. First, it is necessary to determine the size of each sub-image block in the first image and the second image (or called the dimension, that is, the number of pixels of each sub-image block). Smaller blocks can obtain a more detailed local contrast enhancement effect, but may appear unnatural; larger blocks are more uniform overall, but may ignore details. Since the number of pixels in the image is fixed, the larger each sub-image block in the image is, the fewer the number of sub-image blocks included in the image (less details); conversely, the smaller each sub-image block in the image is, the more the number of sub-image blocks included in the image (more details).

[0057] The first number and the second number can be determined based on the resolution of the first image, the resolution of the second image, and the size of the defect on the photovoltaic module to be detected. For example, in a high-resolution image with more details, the second number can be made larger, otherwise the processing is too localized; in a low-resolution image with limited local information, the first number can be made smaller. Another example is that if the detail area is small (such as the crack width is only a few pixels), the first number can be made larger to cover a larger local range. And in the first image and the second image, the sizes of all the sub-image blocks are the same, that is, the sizes of all the first sub-image blocks are the same and the sizes of all the second sub-image blocks are the same; this can improve the uniformity of image processing. If the sub-block sizes are inconsistent, it will result in different local enhancement intensities, thus affecting the uniformity of image processing. For example, if the resolution of the image is 512*512 and the size of each sub-image block is 64*64, then the image will be divided into 8 sub-image blocks of the same size.

[0058] By setting appropriate parameters, adjust the overall contrast and brightness of the image to improve the recognizability of the image without losing details. For the detection of photovoltaic panels, this means increasing the contrast and brightness sufficiently to clearly show cracks, stains, or other defects. The enhanced image should be able to clearly distinguish different photovoltaic panel areas and easily identify any potential defects.

[0059] In some embodiments of the present application, perform histogram equalization operations on each sub-image block based on a contrast increase threshold, including: calculating the histogram of each sub-image block and processing each sub-image block based on the contrast increase threshold; calculating the cumulative distribution function of each processed sub-image block; based on the maximum gray level number and the maximum and minimum values of the cumulative distribution function, remap the pixel values in each sub-image block using the cumulative distribution function to enhance the local contrast of the low-frequency information of the first image and the high-frequency information of the second image.

[0060] Perform histogram equalization on each sub-image block in the first image and the second image respectively, including the following steps. Calculate the histogram H(i) and apply the contrast increase threshold C to adjust the entries of the histogram: H'(i) = MAX(H(i), 0); calculate the cumulative distribution function (CDF) of the adjusted histogram: Use the CDF to remap the pixel values of the original image to enhance the local contrast: where L is the possible maximum gray level number, min(CDF) and max(CDF) are the minimum and maximum values of the CDF respectively, and this step makes full use of the dynamic range of the pixel values. Use different CLAHE algorithms for local contrast enhancement of the low-frequency and high-frequency images to improve the overall visibility and recognition ability of the image. Applying the CLAHE algorithm to the Gaussian-blurred image can optimize the local contrast without introducing over-enhanced noise. That is, apply the CLAHE algorithm to process the low-frequency information image to enhance the local contrast, and also apply the CLAHE algorithm to process the high-frequency information image, but by adjusting parameters (such as the contrast increase threshold), perform a greater degree of contrast enhancement on the area of the high-frequency information image to further highlight the details and textures of the image for the display of details such as hidden cracks.

[0061] In some embodiments of the present application, the image processing method further includes: setting a plurality of candidate thresholds and respectively determining the performance of each candidate threshold when performing histogram equalization processing on the sub-image block; determining the contrast increase threshold from the plurality of candidate thresholds based on a second preset index; where the second preset index includes at least one of the performance indexes of performing histogram equalization processing on the sub-image block: signal-to-noise ratio, image contrast, local contrast mean, edge strength, or edge sharpness.

[0062] Similar to the standard value in Gaussian blur processing, the contrast increase threshold can also be optimally selected from several candidate thresholds. The contrast increase threshold determines the enhancement limit of contrast during the histogram equalization process. A higher contrast increase threshold will enhance the contrast but may also introduce noise; a lower value will reduce this effect but may not be sufficient to make the image clear. Initially, a medium contrast increase threshold can be set, such as 2.0 or 3.0, and then adjusted gradually according to the effect. If the image is too rough or noisy, the contrast increase threshold should be reduced; if the contrast enhancement is insufficient, the contrast increase threshold can be appropriately increased. That is, the most suitable candidate threshold can be selected through certain preset metrics and used as the contrast increase threshold for the CLAHE algorithm. For the sake of distinction, the preset metric here is called the second preset metric, and the second preset metric includes at least one of the following metrics: signal-to-noise ratio, image contrast, local contrast mean, edge intensity, or edge clarity. For example, if the contrast increase threshold is too large, the noise power increases and the signal-to-noise ratio will decrease, indicating that the contrast increase threshold needs to be reduced. If the contrast increase threshold is too small, the contrast is insufficient and the improvement effect of the signal-to-noise ratio is not significant, indicating that the contrast increase threshold can be appropriately increased. Contrast is defined as the change range of the image gray value and can be calculated by the following formula: where contrast is the contrast, and Imax and Imin are the maximum and minimum gray values of the image respectively. A low contrast indicates that the contrast increase threshold is set too small and can be appropriately increased; a too high contrast may be due to an overly large contrast increase threshold and needs to be reduced to avoid noise amplification. Another example is that if the local contrast mean is significantly too low, the contrast increase threshold may be too small and needs to be increased; if the local contrast mean is too high but the noise is significant at the same time, the contrast increase threshold may be too large and needs to be reduced. Another example is that when the edge intensity is insufficient, the contrast increase threshold is increased; when the edge intensity is too large, the contrast increase threshold is reduced. Another example is that when the edge clarity is low, the contrast increase threshold is increased; when the edge clarity is high, the contrast increase threshold is reduced.

[0063] In some embodiments of the present application, the image processing method further includes: reducing the first quantity to reduce the contrast of each first sub-image block and uniformly reflect the low-frequency information of the first image; increasing the second quantity to enhance the contrast of each second sub-image block and prominently reflect the high-frequency information of the second image.

[0064] For the low-frequency information of the photovoltaic module in the image, the first quantity can be reduced to decrease the number of the first sub-image blocks in the first image, thereby reducing the contrast of each first sub-image block and enabling the low-frequency information of the first image to be evenly reflected. Vice versa, for the high-frequency information of the photovoltaic module in the image, the second quantity can be increased to increase the number of the second sub-image blocks in the second image, thereby enhancing the contrast of each second sub-image block and enabling the high-frequency information of the second image to be prominently reflected. In this way, the detailed parts (i.e., the defective parts on the photovoltaic module) can be further prominently displayed, and the other parts can be displayed naturally and evenly, making the image harmonious in terms of perception.

[0065] Based on a similar technical concept, the present application discloses a photovoltaic detection robot. Figure 2 It is a structural example diagram of a photovoltaic detection robot provided by an embodiment of the present application. Please refer to Figure 2 , the photovoltaic detection robot 200 includes: an image acquisition module 210 configured to acquire an initial image of a photovoltaic module 230 to be detected; a processor 220 configured to receive the initial image and apply the image processing method disclosed in the above embodiments to optimize the initial image. The photovoltaic detection robot 200 performs electroluminescence detection to enable the image acquisition module to acquire the initial image of the photovoltaic module 230 to be detected and transmit the initial image to the processor 220, and the processor 220 optimizes the initial image, thereby improving the detection efficiency.

[0066] In the above embodiments, the descriptions of the respective embodiments have their own focuses. For parts not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments. In addition, the above embodiments can be freely combined as needed.

Claims

1. An image processing method, characterized in that: For optimizing an initial image of a photovoltaic component to be detected, the initial image is an image of the photovoltaic component to be detected during electroluminescence detection, the method comprising: Performing smoothing on the initial image to obtain a first image of the photovoltaic assembly to be detected, wherein the first image includes low-frequency information of the initial image; Performing high-pass filtering on the initial image based on the first image to obtain a second image of the photovoltaic assembly to be inspected, wherein the second image includes high-frequency information of the initial image; Splitting the first image and the second image into a plurality of sub-image blocks respectively, and performing a histogram equalization operation on each of the sub-image blocks based on a contrast increase threshold, wherein the first image includes a first sub-image block, the second image includes a second sub-image block, and a contrast increase amount of the second sub-image block is greater than a contrast increase amount of the first sub-image block; A weight value is determined according to information of the initial image, and a third image and a fourth image are synthesized based on the weight value to highlight the details of the photovoltaic component to be detected, wherein the third image is an image obtained after the histogram equalization operation is completed on the first image, and the fourth image is an image obtained after the histogram equalization operation is completed on the second image.

2. The image processing method according to claim 1, characterized in that: The initial image includes a first pixel; The step of smoothing the initial image to obtain the first image of the photovoltaic assembly to be detected includes: Performing Gaussian filtering on the initial image; Determine a first pixel region based on a preset radius with the first pixel as the center; A convolution kernel with a preset standard deviation is used to perform Gaussian blur processing on the first pixel domain to obtain the first image after smoothing.

3. The image processing method according to claim 2, characterized in that: Also includes: Setting a plurality of standard deviations to be selected, and respectively determining the performance of each of the standard deviations to be selected when Gaussian blur processing is performed on the first pixel domain; Determining the preset standard deviation from the plurality of standard deviations to be selected based on a first preset indicator; Among them, the first preset indicator includes at least one of the performance indicators of Gaussian blur processing on the first pixel domain: signal-to-noise ratio, structural similarity of the photovoltaic components to be detected, edge preservation or mean square error of the photovoltaic components to be detected.

4. The image processing method according to claim 1, characterized in that: The initial image further includes a second pixel, the first image includes a third pixel, and the high-pass filtering of the initial image based on the first image to obtain the second image of the photovoltaic assembly to be detected includes: The pixel value of a fourth pixel is determined by subtracting the pixel value of the third pixel from the pixel value of the second pixel, and the second image includes the fourth pixel.

5. The image processing method according to claim 1, characterized in that: The first image includes a first number of first sub-image blocks, and the second image includes a second number of second sub-image blocks, wherein the first number is smaller than the second number; The step of splitting the first image and the second image into a plurality of sub-image blocks respectively comprises: determining the first number and the second number based on an image resolution and a size of a target area, wherein the image resolution includes an image resolution of the first image and an image resolution of the second image, and the size of the target area is a defect size on the photovoltaic module to be inspected; Based on the first number, split the first image into the first number of first sub-image blocks, each of which has the same size; Based on the second number, the second image is split into the second number of second sub-image blocks, and the sizes of the second sub-image blocks are the same.

6. The image processing method according to claim 5, characterized in that: The performing a histogram equalization operation on each of the sub-image blocks based on the contrast increase threshold comprises: Calculating a histogram of each of the sub-image blocks, and processing each of the sub-image blocks based on the contrast increase threshold; Calculating the cumulative distribution function of each of the processed sub-image blocks; Based on the maximum grayscale level and the maximum and minimum values ​​of the cumulative distribution function, the pixel values ​​in each of the sub-image blocks are remapped using the cumulative distribution function to enhance the local contrast of the low-frequency information of the first image and the high-frequency information of the second image.

7. The image processing method according to claim 6, characterized in that: Also includes: Setting a plurality of candidate thresholds, and respectively determining the performance of each candidate threshold when performing histogram equalization processing on the sub-image block; Determining the contrast increase threshold value among the plurality of threshold values ​​to be selected based on a second preset indicator; The second preset indicator includes at least one of the performance indicators of the histogram equalization processing performed on the sub-image block: signal-to-noise ratio, image contrast, local contrast mean, edge strength or edge clarity.

8. The image processing method according to claim 6, characterized in that: Also includes: reducing the first number to reduce the contrast of each of the first sub-image blocks so that the low-frequency information of the first image is evenly reflected; The second number is increased to enhance the contrast of each of the second sub-image blocks, so that the high-frequency information of the second image is highlighted.

9. The image processing method according to claim 1, characterized in that: The step of determining a weight value according to information of the initial image, and synthesizing the third image and the fourth image based on the weight value so as to highlight the detail portion of the photovoltaic assembly to be detected includes: determining a first area and a second area based on information of the initial image; The weight value of the first image is increased in the first area, and the weight value of the second image is increased in the second area, so as to highlight the border and detail distribution of the photovoltaic component to be inspected.

10. A photovoltaic inspection robot, characterized in that: include: An image acquisition module is configured to acquire an initial image of the photovoltaic module to be inspected; A processor is configured to receive the initial image and apply the image processing method according to any one of claims 1 to 9 to optimize the initial image.

Images