Method for remotely monitoring state of power equipment of transformer substation based on image analysis

Abstract

The invention belongs to the technical field of image processing, and particularly relates to a substation power equipment state remote monitoring method based on image analysis, which comprises the following steps of: dividing a monitoring image of power equipment of a substation into a plurality of areas with the same size, and dividing each area into four neighborhood image blocks and one middle image block, the side length of a blurring kernel is adjusted through multiple times of iteration, and the target size of the blurring kernel of the neighborhood image block is determined, and the target size of the blurring kernel of the neighborhood image block is determined according to the change speed of the noise degree and the attenuation degree of the change speed of the noise degree during each time of iteration adjustment; determining the side length of the blurring kernel after the next iteration adjustment according to the noise degree of the neighborhood image block after each iteration adjustment and the side length of the blurring kernel; and taking the mean value of the target sizes of the blurring kernels of the neighborhood image blocks as the target size of the blurring kernel of the middle image block. According to the invention, the efficiency of the whole denoising process is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing. More specifically, the present invention relates to a remote monitoring method for the status of substation power equipment based on image analysis. Background Art

[0002] With the rapid development of the power system, the substation, as a key node in power transmission, its safe and stable operation is of crucial importance.

[0003] Remote status monitoring of power equipment is an important means to ensure the reliable operation of substations. By deploying various sensors and monitoring devices in substations, the operation data and status information of power equipment can be collected in real time to achieve remote monitoring and diagnosis.

[0004] However, the quality of monitoring images has a direct impact on the accuracy of inspection results. In practical applications, due to environmental factors (such as bad weather, light changes), equipment performance limitations (such as equipment aging, equipment overheating, lens distortion), and transmission interference (such as signal attenuation, electromagnetic interference), etc., the collected monitoring images often contain noise.

[0005] These noises will reduce the clarity of the images, obscure the detailed features of the equipment, making it difficult for monitoring personnel to accurately judge the status of the equipment, and may even lead to potential safety hazards due to misjudgment; therefore, improving the quality of monitoring images and effectively removing noise is of great significance for ensuring the accuracy and reliability of remote monitoring of the status of substation power equipment.

[0006] As an existing denoising method, Wiener filtering usually adopts a strategy of iteratively adjusting the size of the blur kernel, and the adjustment amount for each iterative adjustment is the same. Therefore, it is necessary to calculate the blur kernel and the denoised image multiple times, resulting in a huge amount of calculation, significantly increasing the processing time, and affecting the real-time performance and efficiency. Summary of the Invention

[0007] To solve the technical problem that the adjustment amount in each iterative adjustment of the above Wiener filtering is the same, resulting in a large amount of calculation and affecting the real-time performance and efficiency, the present invention provides a remote monitoring method for the status of substation power equipment based on image analysis, including: dividing the monitoring image of the power equipment in the substation into multiple regions of the same size; dividing each region into 4 neighborhood image blocks of the same size and 1 region whose size is equal to The intermediate image block; denoise each image block in the region through Wiener filtering: determine the target size of the blur kernel of the neighborhood image block by adjusting the side length of the blur kernel through multiple iterations, and take the average value of the target sizes of the blur kernels of the 4 neighborhood image blocks as the target size of the blur kernel of the intermediate image block; the multiple iterations of adjusting the side length of the blur kernel include: calculating the noise level of the neighborhood image block after each iteration of adjustment according to the high-frequency coefficients in the DCT coefficient matrix obtained by performing discrete cosine transform on the neighborhood image block after each iteration of adjustment; calculating the change speed of the noise level during each iteration of adjustment according to the difference between the noise level of the neighborhood image block after each iteration of adjustment and the initial neighborhood image block, and the side length of the blur kernel after each iteration of adjustment; calculating the attenuation speed of the change speed of the noise level during each iteration of adjustment according to the change speed of the noise level during each iteration of adjustment and its previous iteration of adjustment; determining the side length of the blur kernel after the next iteration of adjustment according to the change speed of the noise level during each iteration of adjustment, the attenuation degree of the change speed of the noise level, and the noise level and the side length of the blur kernel of the neighborhood image block after each iteration of adjustment.

[0008] The present invention takes the average value of the target sizes of the blur kernels of the 4 neighborhood image blocks as the target size of the blur kernel of the intermediate image block, reduces the number of image blocks that need to be iteratively adjusted, predicts the optimal blur kernel size by analyzing the noise level and its change speed and attenuation speed, reduces the number of iterations, and thus speeds up the iterative adjustment speed; in summary, the present invention significantly improves the efficiency of the entire denoising process by reducing the number of image blocks that need to be iteratively adjusted and speeding up the iterative adjustment speed, making it more suitable for the state monitoring scenario of substation power equipment with high requirements for timeliness.

[0009] Preferably, it is characterized in that dividing the monitoring image into multiple regions with the same size includes: dividing the monitoring image into multiple regions, requiring that the sizes of all regions are the same and are equal to , The value range of and is an integer.

[0010] Preferably, the dividing each region into 4 neighborhood image blocks with the same size and 1 intermediate image block with a size equal to includes: the 4 neighborhood image blocks with the same size are respectively: the upper neighborhood image block, the lower neighborhood image block, the left neighborhood image block, and the right neighborhood image block, and the sizes of the upper neighborhood image block and the lower neighborhood image block are equal to , and the sizes of the neighborhood image block and the right neighborhood image block are equal to .

[0011] In the present invention, by dividing the monitored image into multiple regions, and further dividing each region into 4 neighborhood image blocks of the same size and 1 intermediate image block with a size of 5×5, while retaining the local features of the image, the number of image blocks for which the target size of the blur kernel needs to be determined through iterative adjustment is reduced, thereby reducing the computational amount and improving the processing efficiency.

[0012] Preferably, the calculation method for the noise level of the iteratively adjusted neighborhood image block / initial neighborhood image block is: performing a discrete cosine transform on the iteratively adjusted neighborhood image block / initial neighborhood image block to obtain the DCT coefficient matrix of the iteratively adjusted neighborhood image block / initial neighborhood image block; ; where is the noise level of the iteratively adjusted neighborhood image block / initial neighborhood image block, is the size of the iteratively adjusted neighborhood image block / initial neighborhood image block, is the coordinate of the DCT coefficient in the DCT coefficient matrix, represents the coordinate of the DCT coefficient, represents rounding up.

[0013] Preferably, calculating the change rate of the noise level during each iterative adjustment includes: calculating the noise level of the neighborhood image block after the th iterative adjustment and the noise level of the initial neighborhood image block ; calculating the difference ; calculating the ratio of the difference to the side length of the blur kernel after the th iterative adjustment as the change rate of the noise level during the

[0014] th iterative adjustment.

[0015] By calculating the change rate of the noise level, the present invention can dynamically monitor the change trend of the noise during the iteration process, providing a basis for subsequent adjustment of the side length of the blur kernel to ensure that each adjustment can effectively reduce the noise.

[0015] Preferably, calculating the attenuation rate of the change rate of the noise level during each iterative adjustment includes: calculating the difference between the change rate of the noise level during the th iterative adjustment and the change rate of the noise level during the th iterative adjustment as the attenuation rate of the change rate of the noise level during the th iterative adjustment.

[0016] By calculating the attenuation rate, the present invention can predict the convergence trend of noise change, thereby determining when to stop iteration, providing a more accurate basis for adjusting the side length of the blur kernel, and avoiding over-adjustment.

[0017] Preferably, determining the side length of the blur kernel after the next iterative adjustment includes: if is greater than or equal to 0, stop the iteration, and use the size of the blur kernel after the -th iterative adjustment as the target size of the blur kernel for the neighborhood image block; if is less than 0, then calculate the side length of the blur kernel after the -th iterative adjustment : ; where is the side length of the blur kernel after the -th iterative adjustment, is the noise level of the neighborhood image block after the -th iterative adjustment, is the iteration stop parameter, is the change rate of the noise level during the -th iterative adjustment, is the attenuation degree of the change rate of the noise level during the -th iterative adjustment, represents rounding up.

[0018] The present invention combines the noise level of the neighborhood image block after the previous iterative adjustment, the change rate of the noise level during the previous iterative adjustment, and the attenuation degree of the change rate of the noise level to accurately predict the optimal blur kernel size, reduce unnecessary iterations, and significantly improve the processing efficiency.

[0019] Preferably, when determining the target size of the blur kernel for the neighborhood image block by adjusting the side length of the blur kernel through multiple iterations, it further includes: initially, set an initial blur kernel: if the neighborhood image block is an upper neighborhood image block or a lower neighborhood image block, then set the size of the initial blur kernel to , if the neighborhood image block is a left neighborhood image block or a right neighborhood image block, then set the size of the initial blur kernel to , is the initial side length of the blur kernel, represents rounding up; during the first iterative adjustment, increase the initial side length of the initial blur kernel by 1.

[0020] Preferably, when determining the target size of the blur kernel for the neighborhood image block by adjusting the side length of the blur kernel through multiple iterations, it further includes: when the The change rate of the noise level of the neighborhood image block after the th iteration adjustment is greater than or equal to 0, or the ratio of the noise level of the neighborhood image block after the th iteration adjustment to the noise level of the neighborhood image block after the th iteration adjustment is greater than . Stop the iteration, and use the size of the blurred kernel after the th iteration adjustment as the target size of the blurred kernel of the neighborhood image block.

[0021] Preferably, the denoising process for each image block in the region by Wiener filtering includes: setting the target blurred kernel of the image block according to the target size of the blurred kernel of the image block; applying the target blurred kernel to the Wiener filtering formula, and using the Wiener filtering formula to denoise the image block to obtain the denoised image block.

[0022] The beneficial effects of the present invention are as follows: The present invention uses the mean value of the target sizes of the blurred kernels of 4 neighborhood image blocks as the target size of the blurred kernel of the intermediate image block, reducing the number of image blocks that need to be iteratively adjusted. By analyzing the noise level, its change rate, and attenuation rate, the optimal blurred kernel size is predicted, reducing the number of iterations, and thus accelerating the iterative adjustment speed. In summary, the present invention significantly improves the efficiency of the entire denoising process by reducing the number of image blocks that need to be iteratively adjusted and accelerating the iterative adjustment speed, making it more suitable for the state monitoring scenario of substation power equipment with high timeliness requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flowchart schematically showing the method for remote monitoring of the state of substation power equipment based on image analysis in the present invention; Figure 2 is a schematic diagram schematically showing dividing each region into 4 neighborhood image blocks of the same size and 1 intermediate image block with a size equal to 5×5; Figure 3 is a schematic diagram schematically showing the inspection image containing noise; Figure 4 is a schematic diagram schematically showing the denoised monitoring image obtained by performing Wiener filtering on Figure 3 through the method of this embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0024] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the protection scope of the present invention.

[0025] The following will describe in detail the specific implementation manners of the present invention in conjunction with the accompanying drawings.

[0026] The embodiment of the present invention discloses a method for remotely monitoring the state of substation power equipment based on image analysis. Referring to Figure 1 , it includes steps S1 to S3: S1. Collect monitoring images of the power equipment in the substation.

[0027] Specifically, fixed cameras are installed around key equipment and important areas in the substation, including but not limited to transformers, circuit breakers, instrument transformers, etc.; the monitoring images of the power equipment are collected through the fixed cameras.

[0028] It should be noted that in actual applications, due to environmental factors (such as bad weather, light changes), equipment performance limitations (such as equipment aging, equipment overheating, lens distortion), and transmission interference (such as signal attenuation, electromagnetic interference), etc., the collected monitoring images often have noise, reducing the clarity of the monitoring images, covering up the detailed features of the equipment, and affecting the judgment of the state of the equipment by the monitoring personnel.

[0029] Furthermore, it should be noted that Wiener filtering is a commonly used denoising method. Among them, the blur kernel is the key factor affecting the blur effect. Therefore, usually, the method of iteratively adjusting the size of the blur kernel is adopted. However, in the conventional Wiener filtering, the adjustment amount in each iterative adjustment is the same, which results in the need to calculate the blur kernel and the denoised image multiple times to find the optimal blur kernel size, significantly increasing the processing time and affecting the real-time performance and efficiency.

[0030] S2. Divide the monitoring image into multiple regions with the same size, and divide each region into 4 neighboring image blocks with the same size and 1 intermediate image block with a size equal to .

[0031] It should be noted that, on the one hand, in this embodiment, by dividing the monitoring image into regions and image blocks, and determining the target size of the blur kernel of the neighboring image blocks through multiple iterative adjustments, and determining the target size of the blur kernel of the intermediate image block according to the target sizes of the blur kernels of the 4 neighboring image blocks, the number of image blocks that need to be iteratively adjusted is reduced, which can significantly improve the efficiency of the entire denoising process.

[0032] Specifically, the monitored image is divided into multiple regions, and it is required that the sizes of all regions are the same and equal to , The value range of and is an integer.

[0033] Furthermore, each region is divided into 4 neighborhood image blocks of the same size and 1 middle image block with a size equal to . Among them, the 4 neighborhood image blocks of the same size are respectively: the upper neighborhood image block, the lower neighborhood image block, the left neighborhood image block, and the right neighborhood image block, and the sizes of the upper neighborhood image block and the lower neighborhood image block are equal to , and the sizes of the left neighborhood image block and the right neighborhood image block are equal to .

[0034] It should be noted that the monitored image of the power equipment in the substation is divided into multiple regions of the same size for subsequent unified processing and analysis; each region is further divided into 4 neighborhood image blocks of the same size and 1 middle image block with a size of 5×5, and when determining the target size of the blur kernel of the image block subsequently, the average value of the target sizes of the blur kernels of the 4 neighborhood image blocks is used as the target size of the blur kernel of the middle image block. This division method can retain the local features of the image while reducing the number of image blocks for which the target size of the blur kernel needs to be iteratively adjusted, thereby reducing the computational amount and improving the processing efficiency.

[0035] In one embodiment, the schematic diagram of dividing each region into 4 neighborhood image blocks of the same size and 1 middle image block with a size equal to is as shown in Figure 2 (1) below. Among them, the upper-right vertex of the upper neighborhood image block has the same coordinates as the upper-left vertex of the right neighborhood image block, the lower-right vertex of the right neighborhood image block has the same coordinates as the upper-right vertex of the lower neighborhood image block, and the lower-left vertex of the lower neighborhood image block has the same coordinates as the lower-right vertex of the left neighborhood image block; in another embodiment, the schematic diagram of dividing each region into 4 neighborhood image blocks of the same size and 1 middle image block with a size equal to is as shown in Figure 2 (2) below. Among them, the lower-right vertex of the upper neighborhood image block has the same coordinates as the upper-right vertex of the right neighborhood image block, the lower-left vertex of the right neighborhood image block has the same coordinates as the lower-right vertex of the lower neighborhood image block, and the upper-left vertex of the lower neighborhood image block has the same coordinates as the lower-left vertex of the left neighborhood image block.

[0036] In addition, in the size of the region / image block, the part before " " represents the length of the side parallel to the horizontal direction, and " The length of the side parallel to the vertical direction is represented by what follows; for example: the size of the region is equal to ," What is in front of " represents the length of the side parallel to the horizontal direction, " What is behind " represents the length of the side parallel to the vertical direction; the sizes of the upper neighboring image block and the lower neighboring image block are equal to ," What is in front of " represents the length of the side parallel to the horizontal direction, " What is behind " represents the length of the side parallel to the vertical direction.

[0037] S3. Divide the monitoring images of the power equipment in the substation into multiple regions of the same size; divide each region into 4 neighboring image blocks of the same size and 1 middle image block with a size equal to ; perform denoising processing on each image block in the region through Wiener filtering: determine the target size of the blur kernel of the neighboring image block by adjusting the side length of the blur kernel through multiple iterations, and use the average value of the target sizes of the blur kernels of the 4 neighboring image blocks as the target size of the blur kernel of the middle image block.

[0038] It should be noted that, on the other hand, in this embodiment, by analyzing the noise level, its change speed, and decay speed, the optimal blur kernel size of the neighboring image block is predicted, thereby reducing the number of iterations and accelerating the iterative adjustment speed, which can significantly improve the efficiency of the entire denoising process.

[0039] Specifically, perform denoising processing on each image block in the region through Wiener filtering: determine the target size of the blur kernel of the neighboring image block by adjusting the side length of the blur kernel through multiple iterations; determine the target size of the blur kernel of the middle image block according to the target sizes of the blur kernels of the 4 neighboring image blocks; set the target blur kernel of the neighboring image block / middle image block according to the target size of the blur kernel of the neighboring image block / middle image block; apply the target blur kernel of the neighboring image block / middle image block to the Wiener filtering formula, and use the Wiener filtering formula to perform denoising processing on the neighboring image block / middle image block to obtain the denoised neighboring image block / middle image block.

[0040] Among them, in each region, to determine the target size of the blur kernel of the middle image block according to the target sizes of the blur kernels of the 4 neighboring image blocks, the specific method is: calculate the average value of the target sizes of the blur kernels of the 4 neighboring image blocks, round up the average value of the target sizes of the blur kernels of the 4 neighboring image blocks, and use the result of the rounding up as the target size of the blur kernel of the middle image block.

[0041] Further, for each region, four denoised neighborhood image patches and one denoised intermediate image patch are obtained. Then, the four denoised neighborhood image patches and the one denoised intermediate image patch form the denoised region. All the denoised regions are combined to form the denoised monitoring image.

[0042] Among them, the specific process of iteratively adjusting the side length of the blur kernel multiple times is as follows: 1. Calculate the noise level of the initial neighborhood image patch.

[0043] Specifically, perform a discrete cosine transform on the initial neighborhood image patch to obtain the DCT coefficient matrix of the initial neighborhood image patch. According to the high-frequency coefficients in the DCT coefficient matrix of the initial neighborhood image patch, calculate the noise level of the initial neighborhood image patch and denote it as .

[0044] It should be noted that when using the discrete cosine transform to transform the image patch, the high-frequency coefficients in the DCT coefficient matrix usually reflect the details and noise information in the image. Therefore, they can be used as an index of the noise level, providing a quantitative basis for calculating the subsequent noise change speed.

[0045] 2. Initially, set an initial blur kernel. Apply the initial blur kernel to the Wiener filtering formula, and use the Wiener filtering formula to denoise the initial neighborhood image patch to obtain the denoised neighborhood image patch , calculate the noise level of the denoised neighborhood image patch and denote it as .

[0046] Specifically, if the neighborhood image patch is an upper neighborhood image patch or a lower neighborhood image patch, set the size of the initial blur kernel to , if the neighborhood image patch is a left neighborhood image patch or a right neighborhood image patch, set the size of the initial blur kernel to , is the initial side length of the blur kernel, The specific value of can be set according to the actual application scenario and requirements, and the value range is [3, 7]. In the present invention,

[0047] 3. When performing the first iterative adjustment, increase the initial side length of the initial blur kernel by 1 as the side length of the blur kernel after the first iterative adjustment, and then obtain the blur kernel after the first iterative adjustment. Apply the blur kernel after the first iterative adjustment to the Wiener filtering formula, and use the Wiener filtering formula to denoise the denoised neighborhood image patch to obtain the denoised neighborhood image patch , and denote it as the neighborhood image block after the first iteration adjustment; calculate the noise level of the denoised neighborhood image block and denote it as the noise level of the neighborhood image block after the first iteration adjustment .

[0048] 4. When performing the -th iteration adjustment, , according to the difference between the noise levels of the neighborhood image block after the -th iteration adjustment and the initial neighborhood image block, and the side length of the blur kernel after the -th iteration adjustment, calculate the change rate of the noise level at the -th iteration adjustment; according to the -th iteration adjustment and the change rate of the noise level at the -th iteration adjustment, calculate the attenuation rate of the change rate of the noise level at the -th iteration adjustment; according to the change rate of the noise level and the attenuation degree of the change rate of the noise level at the -th iteration adjustment, and the noise level and the side length of the blur kernel of the neighborhood image block after the -th iteration adjustment, determine the side length of the blur kernel after the -th iteration adjustment, and then obtain the blur kernel after the -th iteration adjustment; apply the blur kernel after the -th iteration adjustment to the Wiener filtering formula, and use the Wiener filtering formula to perform denoising processing on the denoised neighborhood image block to obtain the denoised neighborhood image block , and denote it as the neighborhood image block after the -th iteration adjustment; calculate the noise level of the denoised neighborhood image block and denote it as the noise level of the neighborhood image block after the -th iteration adjustment .

[0049] (1) Among them, perform discrete cosine transform on the neighborhood image block / initial neighborhood image block after iteration adjustment to obtain the DCT coefficient matrix of the neighborhood image block / initial neighborhood image block after iteration adjustment; according to the high-frequency coefficients in the DCT coefficient matrix, calculate the noise level of the neighborhood image block / initial neighborhood image block after iteration adjustment. The specific calculation method is: ; In the formula, is the noise level of the neighborhood image block / initial neighborhood image block after iteration adjustment, is the size of the neighborhood image block / initial neighborhood image block after iteration adjustment, is the coordinate of the DCT coefficient in the DCT coefficient matrix, Denote the DCT coefficient with coordinates ; in the DCT coefficient matrix, Within the range of and Within the range of The DCT coefficients are high-frequency coefficients, where represents rounding up.

[0050] (2) Among them, according to the difference in the noise level between the neighborhood image block adjusted in the th iteration and the initial neighborhood image block, and the side length of the blur kernel adjusted in the th iteration, calculate the change rate of the noise level during the th iteration adjustment, including: calculating the noise level of the neighborhood image block adjusted in the th iteration and the noise level of the initial neighborhood image block The difference ; calculate the difference and the side length of the blur kernel adjusted in the th iteration The ratio , as the change rate of the noise level during the th iteration adjustment .

[0051] It should be specifically noted that when , calculate the change rate of the noise level during the th iteration adjustment, including: calculating the noise level of the denoised neighborhood image block and the noise level of the initial neighborhood image block The difference ; calculate the difference and the initial side length of the blur kernel The ratio , as the change rate of the noise level during the 0th iteration adjustment . It should be noted that as the number of iterations increases, the side length of the blur kernel is constantly increasing, and correspondingly, the noise level of the neighborhood image block is gradually decreasing.

[0052] (3) Among them, according to the adjustment in the

[0053] th iteration and the change rate of the noise level during the th iteration adjustment, calculate the attenuation rate of the change rate of the noise level during the th iteration adjustment, including: calculating the change rate of the noise level during the th iteration adjustment The change rate of the noise level during the The difference from the change rate of the noise level during the th iteration adjustment , as the attenuation rate of the change rate of the noise level during the th iteration adjustment .

[0054] It should be noted that as the number of iterations increases, the change rate of the noise level of the neighborhood image block gradually decreases.

[0055] (4) Among them, according to the change rate of the noise level and the attenuation degree of the change rate of the noise level during the th iteration adjustment, as well as the noise level of the neighborhood image block and the side length of the blur kernel after the th iteration adjustment, determine the side length of the blur kernel after the th iteration adjustment, including: If is greater than or equal to 0, stop the iteration, and use the size of the blur kernel after the th iteration adjustment as the target size of the blur kernel of the neighborhood image block.

[0056] If is less than 0, then calculate the side length of the blur kernel after the th iteration adjustment : ; In the formula, is the side length of the blur kernel after the th iteration adjustment, is the noise level of the neighborhood image block after the th iteration adjustment, is the iteration stop parameter, is the change rate of the noise level during the th iteration adjustment, is the attenuation degree of the change rate of the noise level during the th iteration adjustment, represents rounding up.

[0057] It should be noted that by calculating the attenuation degree of the change rate of the noise level during the th iteration adjustment , predict the convergence trend of the noise change, and then predict the change rate of the noise level during the th iteration adjustment, that is ; Since the condition for iteration to stop is that the change rate of the noise level of the neighborhood image block after the last iteration adjustment is greater than or equal to 0, or the ratio of the noise level of the neighborhood image block after the previous iteration adjustment to the noise level of the neighborhood image block after the last iteration adjustment is greater than , therefore, when predicting the change rate of the noise level during the th iteration adjustment is greater than or equal to 0, stop the iteration and directly use the size of the blur kernel after the th iteration adjustment as the target size of the blur kernel of the neighborhood image block; or when the noise level during the th iteration adjustment is equal to , stop the iteration. During this process, if the noise level changes by , then the change amount of the noise level divided by the predicted change rate of the noise level during the th iteration adjustment is the adjustment amount of the side length of the blur kernel; on the basis of the side length of the blur kernel after the th iteration adjustment, add the adjustment amount to obtain the side length of the blur kernel after the th iteration adjustment, which can achieve the purpose of predicting the optimal blur kernel size and reducing the number of iterations.

[0058] 5. And so on, until the change rate of the noise level of the neighborhood image block after the th iteration adjustment is greater than or equal to 0, or the ratio of the noise level of the neighborhood image block after the th iteration adjustment to the noise level of the neighborhood image block after the th iteration adjustment is greater than , stop the iteration and use the size of the blur kernel after the th iteration adjustment as the target size of the blur kernel of the neighborhood image block, where

[0059] is the iteration stop parameter. Specifically, the value of the iteration stop parameter can be set according to the actual application scenario and requirements, and the value range of the iteration stop parameter is [0.8, 1). In the present invention, the iteration stop parameter

[0060] is set to 0.85. Figure 3 Exemplarily, the schematic diagram of the inspection image containing noise is as shown in Figure 3 . After performing Wiener filtering on Figure 4 through the method of this embodiment, the schematic diagram of the denoised monitoring image obtained is as shown in

[0061] It should be noted that in this embodiment, by analyzing the noise level, its change speed and attenuation speed, the optimal blur kernel size of the neighboring image blocks is predicted, which can reduce the number of iterations and accelerate the iterative adjustment speed. On this basis, directly determining the target size of the blur kernel of the intermediate image block according to the target sizes of the blur kernels of the 4 neighboring image blocks can reduce the number of image blocks that need to be iteratively adjusted; by accelerating the iterative adjustment speed and reducing the number of image blocks that need to be iteratively adjusted, the efficiency of the entire denoising process can be significantly improved, ensuring the timeliness of the condition monitoring of substation power equipment.

Claims

1. A remote monitoring method for the status of substation power equipment based on image analysis, characterized in that Including: Dividing the monitoring images of the power equipment in the substation into multiple regions with the same size; Divide each region into 4 neighborhood image patches of the same size and 1 middle image patch with a size equal to ; perform denoising processing on each image patch in the region through Wiener filtering: determine the target size of the blur kernel of the neighborhood image patches by adjusting the side length of the blur kernel through multiple iterations, and use the mean value of the target sizes of the blur kernels of the 4 neighborhood image patches as the target size of the blur kernel of the middle image patch; The multiple iterative adjustments of the side length of the blur kernel include: calculating the noise level of the neighborhood image block after each iterative adjustment according to the high-frequency coefficients in the DCT coefficient matrix obtained by performing discrete cosine transform on the neighborhood image block after each iterative adjustment; According to the difference between the noise levels of the neighborhood image block after each iterative adjustment and the initial neighborhood image block, and the side length of the blur kernel after each iterative adjustment, calculating the change speed of the noise level during each iterative adjustment; according to the change speed of the noise level during each iterative adjustment and its previous iterative adjustment, calculating the attenuation speed of the change speed of the noise level during each iterative adjustment; according to the change speed of the noise level and the attenuation degree of the change speed of the noise level during each iterative adjustment, and the noise level of the neighborhood image block and the side length of the blur kernel after each iterative adjustment, determining the side length of the blur kernel after the next iterative adjustment.

2. The remote monitoring method for the state of substation power equipment based on image analysis according to claim 1, wherein Dividing the monitoring images into multiple regions with the same size includes: Divide the monitoring image into multiple regions, and require that the sizes of all regions are the same and equal to , The value range of and is an integer.

3. The remote monitoring method for the status of substation electrical equipment based on image analysis according to claim 1, characterized in that, Said dividing each region into 4 neighborhood image patches of the same size and 1 middle image patch with a size equal to comprises: The four neighborhood image patches with the same size are: the upper neighborhood image patch, the lower neighborhood image patch, the left neighborhood image patch, and the right neighborhood image patch, and the sizes of the upper neighborhood image patch and the lower neighborhood image patch are equal to , and the sizes of the neighborhood image patch and the right neighborhood image patch are equal to .

4. The remote monitoring method for the status of substation power equipment based on image analysis according to claim 1, characterized in that, The calculation method for the noise level of the iteratively adjusted neighborhood image block / initial neighborhood image block is: Performing discrete cosine transform on the iteratively adjusted neighborhood image block / initial neighborhood image block to obtain the DCT coefficient matrix of the iteratively adjusted neighborhood image block / initial neighborhood image block; ； In the formula, is the noise level of the iteratively adjusted neighborhood image block / the initial neighborhood image block, is the size of the iteratively adjusted neighborhood image block / the initial neighborhood image block, is the coordinate of the DCT coefficient in the DCT coefficient matrix, represents that the coordinate is the DCT coefficient of, represents rounding up.

5. The remote monitoring method for the state of substation power equipment based on image analysis according to claim 1, characterized in that The calculation of the change speed of the noise level during each iterative adjustment includes: Calculate the noise level of the neighborhood image block after the -th iteration adjustment, and the difference from the noise level of the initial neighborhood image block ; calculate the difference and the ratio to the side length of the blur kernel after the -th iteration adjustment as the change rate of the noise level during the -th iteration adjustment ; .

6. The remote monitoring method for the state of substation power equipment based on image analysis according to claim 1, wherein The calculation of the attenuation speed of the change speed of the noise level during each iterative adjustment includes: Calculate the change rate of the noise level during the th iteration adjustment and the change rate of the noise level during the th iteration adjustment The difference is used as the attenuation rate of the change rate of the noise level during the th iteration adjustment.

7. The remote monitoring method for the status of substation power equipment based on image analysis according to claim 1, characterized in that, The determination of the side length of the blur kernel after the next iterative adjustment includes: If is greater than or equal to 0, stop the iteration, and use the size of the blurred kernel after the -th iteration adjustment as the target size of the blurred kernel for the neighborhood image block; If is less than 0, calculate the side length of the blurred kernel after the th iteration adjustment : ； Wherein, is the side length of the blurred kernel after the -th iterative adjustment, is the noise level of the neighborhood image patch after the -th iterative adjustment, is the iteration stop parameter, is the change rate of the noise level during the -th iterative adjustment, is the attenuation degree of the change rate of the noise level during the -th iterative adjustment, represents rounding up.

8. The remote monitoring method for the status of substation power equipment based on image analysis according to claim 1, characterized in that The determination of the target size of the blur kernel of the neighborhood image block by iteratively adjusting the side length of the blur kernel multiple times further includes: Initially, set an initial blur kernel: if the neighboring image patch is the upper neighboring image patch or the lower neighboring image patch, set the size of the initial blur kernel to , if the neighboring image patch is the left neighboring image patch or the right neighboring image patch, set the size of the initial blur kernel to , is the initial side length of the blur kernel, represents rounding up; During the first iteration adjustment, increase the initial side length of the initial blur kernel by 1.

9. The remote monitoring method for the state of substation power equipment based on image analysis according to claim 1, wherein The determination of the target size of the blur kernel of the neighborhood image block by iteratively adjusting the side length of the blur kernel multiple times further includes: When the change speed of the noise level of the neighborhood image block after the -th iteration adjustment is greater than or equal to 0, or the ratio of the noise level of the neighborhood image block after the -th iteration adjustment to the noise level of the neighborhood image block after the -th iteration adjustment is greater than , stop the iteration, and use the size of the blur kernel after the -th iteration adjustment as the target size of the blur kernel of the neighborhood image block, where is the iteration stop parameter.

10. The remote monitoring method for the state of substation power equipment based on image analysis according to claim 1, wherein, The denoising process of each image block in the region by Wiener filtering includes: Setting the target blur kernel of the image block according to the target size of the blur kernel of the image block; applying the target blur kernel to the Wiener filtering formula, and using the Wiener filtering formula to perform denoising processing on the image block to obtain the denoised image block.

Images