Substation monitoring image noise reduction method and device and electronic equipment

Abstract

The application belongs to the technical field of substation automation, and relates to a substation monitoring image denoising method, device and electronic equipment. The substation monitoring image denoising method comprises the following steps: obtaining a to-be-processed substation monitoring image, and performing texture separation on the to-be-processed substation monitoring image to obtain a texture feature; converting the texture feature to a frequency domain space to determine a high-frequency component cluster and a low-frequency component cluster in the frequency domain space; respectively performing sparse coding on the high-frequency component cluster and the low-frequency component cluster to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix; projecting the high-frequency sparse coding array and the low-frequency sparse coding matrix into a rank space to obtain corresponding high-frequency rank vector distribution and low-frequency rank vector distribution; and determining noise in the to-be-processed substation monitoring image according to the high-frequency rank vector distribution and the low-frequency rank vector distribution and removing the noise, thereby improving the noise removal effect.

Description

Technical Field

[0001] This application belongs to the field of substation automation technology, and in particular relates to a method, device and electronic equipment for noise reduction of substation monitoring images. Background Technology

[0002] With the rapid development of substation automation technology, the functions of integrated substation automation systems are becoming increasingly sophisticated, primarily used to monitor the production and operation status of substations. However, in practical applications, although the system can achieve real-time monitoring of various operating parameters of the substation, it still has shortcomings in ensuring the operating environment of equipment within the station. Particularly in video surveillance, due to the complexity and diversity of the substation environment, monitoring videos are often interfered with by various noises, such as electromagnetic noise and mechanical noise, leading to a decline in video quality and an inability to clearly display equipment operating status and personnel operations, posing a significant challenge to the safe production and operation and maintenance management of substations.

[0003] Existing substation video surveillance systems typically employ traditional image processing methods for video noise reduction. However, these methods often suffer from the following technical shortcomings: firstly, the noise reduction effect is unsatisfactory, failing to effectively remove noise from the video; secondly, the processing speed is slow, unable to meet the needs of real-time monitoring; and thirdly, they lack adaptability, failing to adapt to the noise characteristics of different substation environments. These technical shortcomings limit the application effectiveness of substation video surveillance systems in safe production and operation and maintenance management. Summary of the Invention

[0004] The purpose of this application is to propose a method, device, and electronic equipment for noise reduction of substation monitoring images, which can at least solve or alleviate one of the technical problems existing in the prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a method for denoising substation monitoring images, comprising:

[0007] Acquire monitoring images of substations to be processed, and perform texture separation on the monitoring images of substations to be processed to obtain texture features;

[0008] The texture features are transformed into the frequency domain space to determine high-frequency component clusters and low-frequency component clusters in the frequency domain space.

[0009] Sparse coding is performed on the high-frequency component cluster and the low-frequency component cluster respectively to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix.

[0010] Projecting the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space yields the corresponding high-frequency rank vector distribution and low-frequency rank vector distribution.

[0011] Based on the high-frequency rank vector distribution and the low-frequency rank vector distribution, the noise in the substation monitoring image to be processed is determined and removed.

[0012] In a second aspect, the present invention provides a noise reduction device for substation monitoring images, comprising:

[0013] The texture separation unit is used to acquire the substation monitoring image to be processed and to perform texture separation on the substation monitoring image to obtain texture features;

[0014] A texture conversion unit is used to convert the texture features to a frequency domain space to determine high-frequency component clusters and low-frequency component clusters in the frequency domain space.

[0015] A sparse coding unit is used to perform sparse coding on the high-frequency component cluster and the low-frequency component cluster respectively to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix.

[0016] The projection unit is used to project the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space to obtain the corresponding high-frequency rank vector distribution and low-frequency rank vector distribution.

[0017] The noise removal unit is used to determine and remove noise in the substation monitoring image to be processed based on the high-frequency rank vector distribution and the low-frequency rank vector distribution.

[0018] A third aspect of the present invention provides an electronic device comprising:

[0019] One or more processors;

[0020] Computer-readable media, configured to store one or more programs,

[0021] When the one or more programs are executed by the one or more processors, the one or more processors perform the method as described in any embodiment.

[0022] In this application, texture separation is used to distinguish texture information from structural information in the video, facilitating noise identification in the frequency domain. Furthermore, frequency domain analysis transforms the image or video signal from the spatial domain to the frequency domain, making it easier to identify and process high-frequency noise components. Further, by sparsely coding the high-frequency and low-frequency component clusters, the information most representative of the image or video features can be extracted, while redundancy and noise are removed. Sparse coding captures the main features of the data using a small number of non-zero elements, preserving key information while removing noise in video denoising. Finally, projecting the sparse coding array and matrix into the rank space yields the corresponding high-frequency and low-frequency rank vector distributions. Based on the rank space projection, key information is further extracted, reducing data dimensionality and making it easier to distinguish noise from useful information. According to the high-frequency and low-frequency rank vector distributions, the specific location and characteristics of noise in the video can be determined, effectively removing noise while preserving key information and details. Finally, because it is based on sparse coding and rank space projection, it can adapt to the noise characteristics of different substation environments and ensure that the real-time monitoring needs are met, thus ensuring that the safe production and operation and maintenance management of substations are supported in a timely and effective manner. Attached Figure Description

[0023] The following sections will describe some specific embodiments of the present application in a detailed manner by way of example and not limitation, with reference to the accompanying drawings. The same reference numerals in the drawings denote the same or similar parts or components. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings:

[0024] Figure 1 This is a flowchart illustrating a method for denoising substation monitoring images according to an embodiment of this application;

[0025] Figure 2 This application provides a schematic diagram of the structure of a substation monitoring image noise reduction device according to an embodiment of the present application;

[0026] Figure 3 This is a schematic diagram of the electronic device in this embodiment;

[0027] Figure 4 This is the hardware structure of the electronic device in this embodiment. Detailed Implementation

[0028] To enable those skilled in the art to better understand the technical solutions in the embodiments of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of this application.

[0029] In this application, texture separation is used to distinguish texture information from structural information in the video, facilitating noise identification in the frequency domain. Furthermore, frequency domain analysis transforms the image or video signal from the spatial domain to the frequency domain, making it easier to identify and process high-frequency noise components. Further, by sparsely coding the high-frequency and low-frequency component clusters, the information most representative of the image or video features can be extracted, while redundancy and noise are removed. Sparse coding captures the main features of the data using a small number of non-zero elements, preserving key information while removing noise in video denoising. Finally, projecting the sparse coding array and matrix into the rank space yields the corresponding high-frequency and low-frequency rank vector distributions. Based on the rank space projection, key information is further extracted, reducing data dimensionality and making it easier to distinguish noise from useful information. According to the high-frequency and low-frequency rank vector distributions, the specific location and characteristics of noise in the video can be determined, effectively removing noise while preserving key information and details. Finally, because it is based on sparse coding and rank space projection, it can adapt to the noise characteristics of different substation environments and ensure that the real-time monitoring needs are met, thus ensuring that the safe production and operation and maintenance management of substations are supported in a timely and effective manner.

[0030] Figure 1 This is a flowchart illustrating a substation monitoring image noise reduction method according to an embodiment of this application. Figure 1 As shown, it includes:

[0031] S101. Obtain the monitoring image of the substation to be processed, and perform texture separation on the monitoring image of the substation to be processed to obtain texture features;

[0032] In this embodiment, for example, the substation monitoring images to be processed can be acquired from an image acquisition device. The image acquisition device includes, but is not limited to, cameras, scanners, or other image acquisition devices. Alternatively, the images can be obtained from an image database, such as through direct download.

[0033] In substation monitoring images, texture information may contain crucial information about equipment surface condition, stains, wear, etc., but it may also contain noise. Therefore, separating texture information from structural information (such as edges, shapes, etc.) is essential for subsequent image denoising and enhancement. Optionally, in this embodiment, the texture separation of the substation monitoring image to be processed to obtain texture features includes:

[0034] Obtain the R channel value, G channel value, and B channel value of each pixel in the monitoring image of the substation to be processed;

[0035] The R channel value, G channel value, and B channel value are converted into luminance channel value and chrominance channel value;

[0036] Texture features are obtained by performing texture separation on the substation monitoring image to be processed based on the luminance channel value and color difference channel value corresponding to all pixels.

[0037] The following is an example code for the above-mentioned technical processing:

[0038] import cv2

[0039] import numpy as np

[0040] def get_channel_values(image):

[0041] #Get the R channel value, G channel value, and B channel value of each pixel

[0042] r_channel = image[:,:,0]

[0043] g_channel = image[:,:,1]

[0044] b_channel = image[:,:,2]

[0045] return r_channel,g_channel,b_channel

[0046] def convert_to_yuv(image):

[0047] # Convert RGB image to YUV image

[0048] yuv_image=cv2.cvtColor(image,cv2.COLOR_BGR2YUV)

[0049] #Get luminance channel value and chrominance channel value

[0050] y_channel = yuv_image[:,:,0]

[0051] u_channel = yuv_image[:,:,1]

[0052] v_channel = yuv_image[:,:,2]

[0053] return y_channel,u_channel,v_channel

[0054] def texture_separation(image):

[0055] #Get Channel Value

[0056] r_channel,g_channel,b_channel=get_channel_values(image)

[0057] #Convert to luminance and chrominance channel values

[0058] y_channel,u_channel,v_channel=convert_to_yuv(image)

[0059] # Perform texture separation on the luminance channel values

[0060] # Any texture separation algorithm can be used here, such as wavelet transform, texture filtering, etc.

[0061] #This example demonstrates texture separation using Gaussian filtering.

[0062] texture_y=cv2.GaussianBlur(y_channel,(5,5),0)

[0063] #Merge texture features into the original image's luminance channel

[0064] enhanced_y=cv2.addWeighted(y_channel,0.5,texture_y,0.5,0)

[0065] #Merge texture features into the RGB channels of the original image

[0066] enhanced_image=cv2.merge((enhanced_y,u_channel,v_channel))

[0067] # Convert the image from YUV color space back to RGB color space

[0068] enhanced_image=cv2.cvtColor(enhanced_image,cv2.COLOR_YUV2BGR)

[0069] return enhanced_image

[0070] #Read substation monitoring images to be processed

[0071] image=cv2.imread("image.jpg")

[0072] #Perform texture separation

[0073] result_image=texture_separation(image)

[0074] # Display the result image

[0075] cv2.imshow("Result Image",result_image)

[0076] cv2.waitKey(0)

[0077] cv2.destroyAllWindows()

[0078] Optionally, in this embodiment, the step of performing texture separation on the substation monitoring image to be processed to obtain texture features further includes:

[0079] Obtain the x and y coordinates of each pixel in the monitoring image of the substation to be processed;

[0080] The x-coordinate and y-coordinate of each pixel are projected into an orthogonal space to undergo orthogonal transformation to obtain orthogonal x-coordinate and orthogonal y-coordinate;

[0081] Based on the orthogonal horizontal and vertical coordinates, the texture separation step size is determined;

[0082] The step of performing texture separation on the substation monitoring image to be processed, based on the luminance channel values ​​and chrominance channel values ​​corresponding to all pixels, to obtain texture features includes:

[0083] Based on the texture separation step size, multi-scale features of the luminance channel value and chrominance channel value corresponding to all pixels are statistically analyzed.

[0084] Based on the multi-scale features, texture separation is performed on the substation monitoring image to be processed to obtain texture features. Therefore, the above-mentioned texture separation of the substation monitoring image to obtain texture features has the following technical advantages:

[0085] (1) By projecting the x-coordinate and y-coordinate of a pixel into an orthogonal space and performing an orthogonal transformation, orthogonal x-coordinates and orthogonal y-coordinates can be obtained. Orthogonal transformation can eliminate redundant information in the original coordinate system, making the spatial relationship between pixels clearer, thereby improving the accuracy of texture feature extraction.

[0086] (2) After determining the texture separation step size, multi-scale features of the luminance channel values ​​and chrominance channel values ​​corresponding to all pixels are statistically analyzed based on this step size. Multi-scale features can capture texture details at different scales, thus providing a more comprehensive description of the image's texture information. This texture separation method based on multi-scale features can more effectively extract texture features from substation monitoring images.

[0087] (3) Since there may be various texture types in substation monitoring images, such as equipment surface texture and cable texture, multi-scale feature extraction method can enhance the algorithm's adaptability to different texture types. Both coarse and fine textures can be effectively extracted.

[0088] (4) The automated texture separation process can greatly reduce the need for manual intervention and improve the automation level of image analysis. This is of great significance for the rapid processing and real-time analysis of substation monitoring images, and helps to improve the safety and reliability of substation operation.

[0089] (5) The extracted texture features can serve as the basis for subsequent image processing and analysis, such as object detection and anomaly recognition. By utilizing these texture features, the accuracy and efficiency of image processing algorithms can be further improved.

[0090] Therefore, the exemplary implementation code of the above technical processing is as follows:

[0091]

[0092]

[0093] Optionally, in this embodiment, based on the texture separation step size, multi-scale features of the luminance channel values ​​and chrominance channel values ​​corresponding to all pixels are statistically analyzed, such as: extracting feature values ​​from the luminance channel values ​​and chrominance channel values ​​corresponding to all pixels; and statistically analyzing the multi-scale features of the feature values ​​of the luminance channel values ​​and chrominance channel values ​​corresponding to all pixels based on the texture separation step size.

[0094] Therefore, based on the technical solution provided above, statistically analyzing the multi-scale features of luminance and chrominance channel values ​​during texture separation of substation monitoring images offers the following technical advantages:

[0095] (1) By extracting feature values ​​from the luminance and chrominance channels and combining them with multi-scale statistics, richer texture information can be obtained. This information includes not only the luminance and color information of the pixels themselves, but also the texture structure information at different scales, thus enabling a more comprehensive description of the texture features of the image.

[0096] (2) The extraction of multi-scale features makes the algorithm robust to changes in texture scale. At different scales, textures may exhibit different features. By statistically analyzing feature values ​​at multiple scales, the algorithm can adapt to texture changes at different scales and improve its generalization ability.

[0097] (3) By selecting an appropriate texture separation step size, the computational load can be reduced to some extent, thereby improving the computational efficiency of the algorithm. At the same time, the statistics of multi-scale features can also reduce the interference of redundant information to some extent, making the extracted texture features more accurate and effective.

[0098] (4) The extracted multi-scale texture features can serve as the basis for subsequent image processing and analysis, such as object detection and anomaly recognition. These features can provide more accurate and comprehensive information support for subsequent algorithms, thereby improving the accuracy and efficiency of the algorithms.

[0099] Below is an example implementation code for extracting multi-scale features from luminance and chrominance channel values:

[0100]

[0101] S102. The texture features are converted to the frequency domain space to determine the high-frequency component clusters and low-frequency component clusters in the frequency domain space.

[0102] Optionally, the method further includes: constructing a filter, the filter including a conversion module, a high-frequency filtering module and a low-frequency filtering module, wherein the conversion module is cascaded with the high-frequency filtering module and the low-frequency filtering module respectively.

[0103] The step of converting the texture features to the frequency domain space to determine high-frequency component clusters and low-frequency component clusters in the frequency domain space includes:

[0104] The texture features are input into the conversion module, and the texture features are converted into frequency domain signals and input into the high-frequency filtering module and the low-frequency filtering module;

[0105] The high-frequency filtering module extracts all high-frequency components from the frequency domain signal to form the high-frequency component cluster;

[0106] The low-frequency filtering module extracts all low-frequency components from the frequency domain signal to form the low-frequency component cluster.

[0107] Optionally, the high-frequency filtering module extracts all high-frequency components from the frequency domain signal to form the high-frequency component cluster, including:

[0108] The high-frequency filtering module divides the frequency domain signal into spatial frequency distributions and calculates the entropy of each spatial frequency distribution region to extract the frequency domain signal covered by the spatial frequency distribution region when the entropy is greater than a set entropy threshold, so as to form the high-frequency component cluster.

[0109] Therefore, the frequency domain processing technology and filter construction method involved in the above-mentioned substation monitoring image processing solution have the following technical advantages:

[0110] (1) By converting texture features to the frequency domain for analysis, the system can more effectively identify and separate high-frequency and low-frequency components in an image. In the frequency domain, the details and structure of the texture are clearly reflected in different frequency components, making the analysis more intuitive and accurate.

[0111] (2) Frequency domain processing has a natural ability to suppress noise and redundant information. The high-frequency filtering module can extract high-frequency detail information in the image, while the low-frequency filtering module preserves the image's contour and basic structural information. This helps to eliminate or reduce noise in the image and extract key information useful for subsequent processing.

[0112] (3) By constructing a filter that includes a conversion module, a high-frequency filtering module, and a low-frequency filtering module, the system can achieve fast frequency domain conversion and component extraction of texture features. This modular design makes the processing flow clearer and more efficient, reducing unnecessary computational overhead.

[0113] (4) The high-frequency filtering module and the low-frequency filtering module in the filter can be customized and adjusted according to specific needs. For example, the high-frequency filtering module can be set with different entropy thresholds according to different texture types and application scenarios to achieve more accurate high-frequency component extraction. This flexibility enables the system to adapt to a wider range of monitoring scenarios and application needs.

[0114] (5) By extracting high-frequency component clusters through spatial frequency distribution partitioning and entropy calculation, the system can more accurately identify key texture information in the image. This method is robust to interference factors such as illumination changes and occlusion, ensuring the stability and reliability of the algorithm under different conditions.

[0115] (6) The extracted high-frequency and low-frequency component clusters can provide strong support for subsequent image processing and analysis. For example, the high-frequency component clusters can be used for tasks such as detail enhancement and edge detection, while the low-frequency component clusters can be used for tasks such as image reconstruction and target recognition. This multi-scale information processing approach helps improve the performance and efficiency of the entire monitoring system.

[0116] To this end, an exemplary code for implementing the above scheme is provided below: the process of constructing and frequency domain analysis of the filter is simulated using a Python-based library (such as OpenCV and NumPy), demonstrating how to construct a filter containing a transformation module (Fourier transform), a high-frequency filtering module, and a low-frequency filtering module, and how to use these modules to separate the high-frequency and low-frequency components in texture features.

[0117]

[0118] `low_freq[high_freq!=0]=0` # Sets the high-frequency component to zero to obtain the low-frequency component.

[0119] return low_freq

[0120] # Convert texture features (in this case, an image) into a frequency domain signal.

[0121] fshift=fourier_transform(image)

[0122] #Set entropy threshold

[0123] entropy_threshold = 2

[0124] #Extracting high-frequency components

[0125] high_freq_components=high_frequency_filter(fshift,entropy_threshold)

[0126] #Extracting low-frequency components

[0127] low_freq_components=low_frequency_filter(fshift)

[0128] high_freq_image=np.fft.ifftshift(high_freq_components)

[0129] high_freq_image=np.abs(np.fft.ifft2(high_freq_image))

[0130] high_freq_image=(high_freq_image-np.min(high_freq_image)) /

[0131] (np.max(high_freq_image)-np.min(high_freq_image))*255

[0132] high_freq_image=np.uint8(high_freq_image)

[0133] low_freq_image=np.fft.ifftshift(low_freq_components)

[0134] low_freq_image=np.abs(np.fft.ifft2(low_freq_image))

[0135] low_freq_image=(low_freq_image-np.min(low_freq_image)) /

[0136] (np.max(low_freq_image)-np.min(low_freq_image))*255

[0137] low_freq_image=np.uint8(low_freq_image)

[0138] # Display results

[0139] cv2.imshow('Original Image',image)

[0140] Optionally, the low-frequency filtering module extracts all low-frequency components from the frequency domain signal to form the low-frequency component cluster, including:

[0141] The low-frequency filtering module divides the frequency domain signal into spatial frequency distributions and calculates the entropy of each spatial frequency distribution region to extract the frequency domain signal covered by the spatial frequency distribution region when the entropy is less than a set entropy threshold, so as to form the low-frequency component cluster.

[0142] In this embodiment, by dividing the frequency domain signal into spatial frequency distributions and calculating the entropy of each spatial frequency distribution region, spatial frequency distribution regions with entropy less than a set entropy threshold can be extracted. The frequency domain signals corresponding to these spatial frequency distribution regions represent the low-frequency components in the image, containing the overall structure and background information of the image. Meanwhile, as mentioned above, high-frequency signals typically correspond to high-frequency information such as details and textures in the image, while low-frequency signals correspond to smooth areas and the background, thus ensuring the overall accuracy of noise reduction.

[0143] S103. Perform sparse coding on the high-frequency component cluster and the low-frequency component cluster respectively to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix.

[0144] Optionally, in this embodiment, the step of performing sparse coding on the high-frequency component cluster and the low-frequency component cluster to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix includes:

[0145] The high-frequency component cluster and the low-frequency component cluster are vectorized respectively to obtain the corresponding high-frequency component vector and low-frequency component vector.

[0146] Based on the linear combination vector of the set basis functions, the high-frequency component vector and the low-frequency component vector are sparsely encoded respectively to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix.

[0147] Therefore, in image processing or feature representation tasks, the above-mentioned sparse coding of high-frequency component clusters and low-frequency component clusters achieves effective feature extraction and dimensionality reduction, and has the following technical advantages:

[0148] (1) Through sparse coding, high-frequency and low-frequency component clusters can be transformed into a more compact and robust representation. This representation is more robust to noise and local deformation because only a few basis functions are activated to reconstruct the input data during sparse coding, which makes the representation less sensitive to the specific details of the input data.

[0149] (2) Sparse coding transforms the original high-dimensional data (such as high-frequency component clusters and low-frequency component clusters) into a low-dimensional sparse coding array or matrix. This dimensionality reduction process can greatly reduce the computational load of subsequent processing steps (such as classification, recognition, etc.) and improve the overall computational efficiency of the system.

[0150] (3) Since sparse coding approximates the original data based on a linear combination of basis functions, discriminative basis functions can be selected to enhance the discriminative power of the features. In the sparse coding process of high-frequency and low-frequency component clusters, basis functions that can capture key information in high-frequency and low-frequency features can be selected respectively, thereby enhancing the discriminative power of the encoded features.

[0151] (4) By sparsely coding the high-frequency and low-frequency component clusters separately, two independent sparse coding arrays or matrices can be obtained. These two sparse codes can be easily fused to combine information from the high-frequency and low-frequency features. This fusion method can be achieved through simple concatenation, weighting, or more complex fusion strategies, providing more information input for subsequent processing steps.

[0152] The exemplary code above is as follows:

[0153] import numpy as np

[0154] #Example data for high-frequency component clustering and low-frequency component clustering

[0155] high_freq_clusters=np.array([[1,2,3],[4,5,6],[7,8,9]])

[0156] low_freq_clusters=np.array([[10,20,30],[40,50,60]])

[0157] #Vectorization of high-frequency component clusters and low-frequency component clusters

[0158] high_freq_vector=high_freq_clusters.flatten()

[0159] low_freq_vector=low_freq_clusters.flatten()

[0160] # Define the linear combination vector of basis functions

[0161] basis_functions=np.array([[1,0,0],[0,1,0],[0,0,1]])

[0162] #Sparse encoding of high-frequency component vectors

[0163] high_freq_sparse_code=np.linalg.lstsq(basis_functions,high_freq_vector,rcond=None)[0]

[0164] #Sparse encoding of low-frequency component vectors

[0165] low_freq_sparse_code=np.linalg.lstsq(basis_functions,low_freq_vector,rcond=None)[0]

[0166] print("High-frequency sparse code array:", high_freq_sparse_code)

[0167] print("Low-frequency sparse coding matrix:", low_freq_sparse_code)

[0168] In the code above, example data for high-frequency component clusters and low-frequency component clusters are first defined. Then, the `flatten()` function is used to flatten the cluster data into a one-dimensional vector. Next, a linear combination vector of basis functions, `basis_functions`, is defined, assuming that the basis functions are unit vectors. Finally, the `np.linalg.lstsq()` function is used to sparsely encode the high-frequency and low-frequency vectors, resulting in a high-frequency sparse coding array and a low-frequency sparse coding matrix.

[0169] S104. Project the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space to obtain the corresponding high-frequency rank vector distribution and low-frequency rank vector distribution.

[0170] Optionally, in this embodiment, projecting the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space to obtain the corresponding high-frequency rank vector distribution and low-frequency rank vector distribution includes:

[0171] Projecting the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space yields the corresponding high-frequency rank tensor and low-frequency rank tensor.

[0172] Based on the high-frequency rank tensor and the low-frequency rank tensor, statistically analyze the distribution of the high-frequency rank vector and the distribution of the low-frequency rank vector.

[0173] Therefore, in image processing or feature analysis, projecting the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space to obtain the high-frequency rank vector distribution and the low-frequency rank vector distribution has the following technical advantages:

[0174] (1) By projecting sparse coding arrays and matrices into the rank space, the original high-dimensional data can be compressed into a low-dimensional rank tensor. This dimensionality reduction process not only reduces the storage and transmission costs of the data, but also reduces the computational complexity of subsequent analysis.

[0175] (2) Rank space projection can preserve key structural information in the original data. High-frequency rank tensors mainly reflect details such as edges and textures in the image, while low-frequency rank tensors mainly reflect the overall shape and contour of the image. This preservation of feature structure is helpful for feature extraction and recognition in subsequent tasks.

[0176] (3) By converting the rank tensor into a rank vector distribution, the distribution of high-frequency and low-frequency features in the rank space can be displayed more intuitively. This statistical distribution helps to analyze the overall characteristics and local differences of images or data.

[0177] (3) Since the calculation of rank space projection and statistical distribution is based on the overall data, this method has good robustness to local noise and outliers. Even if there are some noise or outliers in the original data, the rank tensor and rank vector distribution after projection can still remain relatively stable, providing reliable feature representation for subsequent processing.

[0178] (4) By projecting the high-frequency and low-frequency components into the rank space respectively, feature representations at different scales can be obtained. This multi-scale analysis helps to understand the structure and content of images or data more comprehensively, and provides rich feature information for subsequent tasks such as image recognition and object detection.

[0179] Statistical analysis of high-frequency and low-frequency rank vector distributions can be achieved using histograms, for example. In a specific application scenario, such as using Python's Matplotlib library to plot histograms:

[0180] import numpy as np

[0181] import matplotlib.pyplot as plt

[0182] #Example data for high-frequency and low-frequency rank vectors

[0183] high_freq_rank_vectors=np.array([1,2,3,4,5])

[0184] low_freq_rank_vectors=np.array([10,20,30,40,50])

[0185] #Statistical distribution of high-frequency rank vectors

[0186] high_freq_rank_hist,high_freq_rank_bins=np.histogram(high_freq_rank_vectors,bins=10)

[0187] #Statistical distribution of low-frequency rank vectors

[0188] low_freq_rank_hist,low_freq_rank_bins=np.histogram(low_freq_rank_vectors,bins=10)

[0189] #Draw a histogram of high-frequency rank vector distribution

[0190] plt.bar(high_freq_rank_bins[:-1],high_freq_rank_hist,

[0191] width=np.diff(high_freq_rank_bins),align='edge')

[0192] plt.xlabel('High-Frequency Rank Vector')

[0193] plt.ylabel('Frequency')

[0194] plt.title('High-Frequency Rank Vector Distribution Histogram')

[0195] plt.show()

[0196] #Draw a histogram of low-frequency rank vector distribution

[0197] plt.bar(low_freq_rank_bins[:-1],low_freq_rank_hist,width=np.diff(low_freq_rank_bins),align='edge')

[0198] plt.xlabel('Low-frequency rank vector')

[0199] plt.ylabel('Frequency')

[0200] plt.title('Low-frequency rank vector distribution histogram')

[0201] plt.show()

[0202] In the code above, the `np.histogram()` function is used to statistically analyze the distribution of the vectors, where the `bins` parameter specifies the number of bins in the histogram. Additionally, the `plt.bar()` function from the Matplotlib library is used to plot histograms of the high-frequency and low-frequency rank vector distributions.

[0203] S105. Based on the high-frequency rank vector distribution and the low-frequency rank vector distribution, determine and remove the noise in the substation monitoring image to be processed.

[0204] Optionally, determining and removing noise from the substation monitoring image to be processed based on the high-frequency rank vector distribution and the low-frequency rank vector distribution includes:

[0205] Projecting the high-frequency rank vector distribution and the low-frequency rank vector distribution onto Gaussian space yields the corresponding Gaussian distribution;

[0206] The maximum likelihood estimates of the Gaussian distributions corresponding to the high-frequency rank vector distribution and the low-frequency rank vector distribution are determined and compared with the set maximum likelihood threshold to identify and remove noise in the substation monitoring image to be processed.

[0207] Therefore, projecting the high-frequency rank vector distribution and the low-frequency rank vector distribution onto Gaussian space, and using the maximum likelihood estimate of the Gaussian distribution to determine and remove noise in the substation monitoring image, has the following technical advantages:

[0208] (1) By projecting the rank vector distribution onto Gaussian space, the characteristics of Gaussian distribution can be used to accurately detect noise. Ideally, noise-free data will exhibit a certain Gaussian distribution pattern, while noisy data may deviate from this distribution pattern. Therefore, by comparing the maximum likelihood estimate of the Gaussian distribution corresponding to the rank vector distribution with a set threshold, noisy data can be effectively identified.

[0209] (2) This method has a certain degree of adaptability. By setting the maximum likelihood threshold, the sensitivity of noise detection can be adjusted according to different application scenarios and data characteristics. This is of great significance for processing substation monitoring images with different noise levels.

[0210] (3) Since this method is based on the rank vector distribution of high and low frequencies for noise detection, it can preserve the image's detail information while removing noise. High-frequency components typically contain the image's details and texture information, while low-frequency components reflect the image's overall structure and contours. By processing the high-frequency and low-frequency components separately, the image's clarity and detail can be maintained while removing noise.

[0211] Below is a simple example implementation code illustrating how to project a rank vector distribution onto a Gaussian space and perform noise detection:

[0212] import numpy as np

[0213] from scipy.stats import norm

[0214] #High-frequency rank vector and low-frequency rank vector

[0215] high_freq_vectors = np.random.randn(1000) # High-frequency rank vectors

[0216] low_freq_vectors = np.random.randn(1000) # Low-frequency rank vectors

[0217] #Project the rank vector onto Gaussian space

[0218] #Mean and standard deviation of Gaussian distribution

[0219] mean_high=np.mean(high_freq_vectors)

[0220] std_high=np.std(high_freq_vectors)

[0221] mean_low=np.mean(low_freq_vectors)

[0222] std_low=np.std(low_freq_vectors)

[0223] # Calculate the maximum likelihood estimate (for a Gaussian distribution, this is the value of the probability density function).

[0224] def gaussian_likelihood(x,mean,std):

[0225] return norm.pdf(x,mean,std)

[0226] #For high-frequency rank vectors

[0227] high_freq_likelihoods=gaussian_likelihood(high_freq_vectors,mean_high,std_high)

[0228] #Set the maximum likelihood threshold for noise detection

[0229] max_likelihood_threshold = norm.pdf(mean_high,mean_high,std_high)*0.1 # The threshold is 10% of the probability density at the mean.

[0230] #Detect and remove noise

[0231] high_freq_vectors_cleaned=

[0232] high_freq_vectors[high_freq_likelihoods>max_likelihood_threshold]

[0233] For example, projecting the high-frequency rank vector distribution and the low-frequency rank vector distribution into a Gaussian space to obtain the corresponding Gaussian distribution may include: merging the high-frequency rank vector distribution and the low-frequency rank vector distribution into a single matrix, with each vector as a row of the matrix; performing a mean-reduction operation on the merged matrix, i.e., subtracting the mean of the entire dimension from the mean of each dimension; calculating the covariance matrix of the merged matrix; performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; sorting the eigenvectors according to the magnitude of the eigenvalues; selecting the top N eigenvectors as principal components; and multiplying the merged matrix by the selected principal components to obtain the projected data matrix. The projected data matrix is ​​the representation of the high-frequency rank vector distribution and the low-frequency rank vector distribution in Gaussian space, which can be regarded as the corresponding Gaussian distribution.

[0234] Therefore, projecting the high-frequency and low-frequency rank vector distributions into Gaussian space and using principal component analysis (PCA) to obtain their representations in Gaussian space is an effective feature extraction and dimensionality reduction technique with the following technical advantages:

[0235] (1) By merging the high-frequency rank vector and the low-frequency rank vector into a single matrix and applying PCA, we can project the data from the original high-dimensional space into a low-dimensional principal component space. This significantly reduces the dimensionality of the data and decreases the computational and storage requirements.

[0236] (2) PCA is an unsupervised learning method that reduces dimensionality based on the statistical properties of the data (such as the covariance matrix). Since noise often manifests as random fluctuations in the data, PCA may treat these noise components as minor features and ignore them during the projection process, thereby achieving the effect of suppressing noise.

[0237] (3) By selecting the top N principal components, we are actually selecting the most important features in the data. These features usually contain most of the information in the original data, while noise and other irrelevant information are removed or weakened.

[0238] The following exemplary implementation code projects high-frequency and low-frequency rank vector distributions onto a Gaussian space (using PCA dimensionality reduction):

[0239] import numpy as np

[0240] from sklearn.decomposition import PCA

[0241] #High-frequency rank vector and low-frequency rank vector

[0242] high_freq_vectors=np.random.randn(100,10)

[0243] low_freq_vectors=np.random.randn(100,5)

[0244] # Combine the high-frequency rank vector and the low-frequency rank vector into a single matrix.

[0245] # Expand the low-frequency rank vector to the same dimension as the high-frequency rank vector (e.g., by padding with zeros or interpolation).

[0246] low_freq_vectors_padded=np.pad(low_freq_vectors,((0,0),(0,5)),'constant',constant_values=0)

[0247] combined_vectors=np.concatenate((high_freq_vectors,low_freq_vectors_padded),axis=0)

[0248] # Perform PCA dimensionality reduction on the merged matrix

[0249] pca = PCA(n_components = 5) # Assuming the first 5 principal components are selected

[0250] projected_data=pca.fit_transform(combined_vectors)

[0251] The projected data matrix is ​​the representation of the high-frequency rank vector distribution and the low-frequency rank vector distribution in Gaussian space.

[0252] #Showing the data after dimensionality reduction

[0253] print(projected_data)

[0254] #import matplotlib.pyplot as plt

[0255] #plt.scatter(projected_data[:,0],projected_data[:,1])

[0256] #plt.show()

[0257] Please note that `low_freq_vectors_padded` in the code above is a hypothetical padding to expand the dimension of low-frequency rank vectors to be the same as that of high-frequency rank vectors. In practical applications, the appropriate padding or interpolation method can be selected based on the specific data structure and task requirements.

[0258] Optionally, in this embodiment, determining the maximum likelihood estimate of the Gaussian distribution corresponding to the high-frequency rank vector distribution and the low-frequency rank vector distribution may include, for example, calculating the Gaussian distribution probability density function value corresponding to the gradient magnitude of each pixel based on the estimated Gaussian distribution parameters of the high-frequency rank vector and the low-frequency rank vector, obtaining the corresponding maximum likelihood estimate, and comparing it with the set maximum likelihood threshold.

[0259] Therefore, when using the maximum likelihood estimate of a Gaussian distribution to determine and remove noise from an image, especially based on the Gaussian distribution parameters of the high-frequency and low-frequency rank vectors, the following technical advantages are available:

[0260] (1) By calculating the Gaussian probability density function value corresponding to the gradient magnitude of each pixel, a quantitative index can be obtained to measure whether the pixel belongs to noise. This probability-based method can detect noise more accurately, especially when the noise distribution is complex or the noise level is high.

[0261] (2) The distribution of pixel values ​​is described using Gaussian distribution parameters (mean and variance), which makes the model more flexible and adaptable. These parameters can be automatically estimated through data-driven methods without manual setting, thereby improving the automation and robustness of the algorithm.

[0262] (3) The sensitivity of noise detection can be controlled by setting the maximum likelihood threshold. The threshold can be adjusted according to the specific application scenario and requirements to achieve different levels of noise removal effect.

[0263] (4) Since this method is based on the Gaussian distribution of the high-frequency rank vector and the low-frequency rank vector for noise detection, it can preserve the image's detail information while removing noise. The high-frequency components usually contain the image's details and texture information, while the low-frequency components reflect the image's overall structure and contours. By processing the high-frequency and low-frequency components separately, the image's sharpness and detail can be maintained while removing noise.

[0264] Below is an example implementation code for calculating the maximum likelihood estimate of the Gaussian distribution for each pixel and performing noise detection:

[0265]

[0266] # Treat pixels below the threshold as noise and set them to zero.

[0267] denoised_image_gradients=np.where(likelihoods_high>max_likelihood_threshold,image_gradients,0)

[0268] #denoised_image_gradients contains the magnitude of the denoised gradient.

[0269] Figure 2 This is a schematic diagram of the structure of a substation monitoring image noise reduction device according to an embodiment of this application. Figure 2 As shown, it includes:

[0270] The texture separation unit 201 is used to acquire the substation monitoring image to be processed and perform texture separation on the substation monitoring image to obtain texture features.

[0271] Texture conversion unit 202 is used to convert the texture features to the frequency domain space to determine high-frequency component clusters and low-frequency component clusters in the frequency domain space.

[0272] Sparse coding unit 203 is used to perform sparse coding on the high-frequency component cluster and the low-frequency component cluster respectively to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix.

[0273] Projection unit 204 is used to project the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space to obtain the corresponding high-frequency rank vector distribution and low-frequency rank vector distribution.

[0274] The noise removal unit 205 is used to determine and remove noise in the substation monitoring image to be processed based on the high-frequency rank vector distribution and the low-frequency rank vector distribution.

[0275] Figure 3 This is a schematic diagram of the structure of the electronic device in this embodiment; the electronic device may include:

[0276] One or more processors 301;

[0277] Computer-readable medium 302 may be configured to store one or more programs.

[0278] When the one or more programs are executed by the one or more processors, the one or more processors perform the method as described in the above embodiments.

[0279] Figure 4 This is the hardware structure of the electronic device in this embodiment; such as Figure 4 As shown, the hardware structure of the electronic device may include: a processor 401, a communication interface 402, a computer-readable medium 403, and a communication bus 404.

[0280] The processor 401, communication interface 402, and computer-readable medium 403 communicate with each other via communication bus 404.

[0281] Optionally, the communication interface 402 can be an interface of a communication module, such as the interface of a GSM module;

[0282] Specifically, the processor 401 can be configured to: acquire a substation monitoring image to be processed, and perform texture separation on the substation monitoring image to obtain texture features; convert the texture features to the frequency domain space to determine high-frequency component clusters and low-frequency component clusters in the frequency domain space; perform sparse coding on the high-frequency component clusters and the low-frequency component clusters respectively to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix; project the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space to obtain the corresponding high-frequency rank vector distribution and low-frequency rank vector distribution; and determine and remove noise in the substation monitoring image to be processed based on the high-frequency rank vector distribution and the low-frequency rank vector distribution.

[0283] Processor 401 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor.

[0284] The computer-readable medium 403 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0285] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code configured to perform the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access storage media (RAM), read-only storage media (ROM), erasable programmable read-only storage media (EPROM or flash memory), optical fibers, portable compact disk read-only storage media (CD-ROM), optical storage media, magnetic storage media, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program configured for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0286] Computer program code configured to perform the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as "C" or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0287] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions configured to perform a specified logical function. Specific sequences are present in the above specific embodiments, but these sequences are merely exemplary; in actual implementations, these steps may be fewer, more, or executed in a different order. That is, in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0288] In another aspect, this application also provides a computer-readable medium having a computer program stored thereon that, when executed by a processor, implements the methods described in the above embodiments.

[0289] In another aspect, this application also provides a computer-readable medium, which may be included in the apparatus described in the above embodiments; or it may exist independently and not assembled into the apparatus. The computer-readable medium carries one or more programs, which, when executed by the apparatus, cause the apparatus to: acquire a substation monitoring image to be processed, and perform texture separation on the substation monitoring image to obtain texture features; convert the texture features to a frequency domain space to determine high-frequency component clusters and low-frequency component clusters in the frequency domain space; perform sparse coding on the high-frequency component clusters and the low-frequency component clusters respectively to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix; project the high-frequency sparse coding array and the low-frequency sparse coding matrix into a rank space to obtain corresponding high-frequency rank vector distributions and low-frequency rank vector distributions; and determine and remove noise in the substation monitoring image to be processed based on the high-frequency rank vector distributions and low-frequency rank vector distributions.

[0290] The terms "first," "second," "first," or "second" as used in the various embodiments of this disclosure may modify various components regardless of their order and / or importance, but these terms do not limit the corresponding components. The above terms are configured only for the purpose of distinguishing an element from other elements. For example, "first user equipment" and "second user equipment" refer to different user equipments, although both are user equipment. For example, without departing from the scope of this disclosure, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element.

[0291] When a component (e.g., a first component) is referred to as being "(operably or communicatively) coupled" or "(operably or communicatively) coupled to" or "connected to" another component (e.g., a second component), it should be understood that the first component is directly connected to the second component or that the first component is indirectly connected to the second component via yet another component (e.g., a third component). Conversely, it can be understood that when a component (e.g., a first component) is referred to as being "directly connected" or "directly coupled" to another component (the second component), no component (e.g., a third component) is inserted between the two.

[0292] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for denoising substation monitoring images, characterized in that, include: Acquire monitoring images of substations to be processed, and perform texture separation on the monitoring images of substations to be processed to obtain texture features; The texture features are transformed into the frequency domain space to determine high-frequency component clusters and low-frequency component clusters in the frequency domain space. Sparse coding is performed on the high-frequency component cluster and the low-frequency component cluster respectively to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix. Projecting the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space yields the corresponding high-frequency rank vector distribution and low-frequency rank vector distribution. Based on the high-frequency rank vector distribution and the low-frequency rank vector distribution, the noise in the substation monitoring image to be processed is determined and removed, including: projecting the high-frequency rank vector distribution and the low-frequency rank vector distribution onto a Gaussian space to obtain the corresponding Gaussian distribution; The maximum likelihood estimates of the Gaussian distribution corresponding to the high-frequency rank vector distribution and the low-frequency rank vector distribution are determined and compared with the set maximum likelihood threshold to identify and remove noise in the substation monitoring image to be processed.

2. The method according to claim 1, characterized in that, The process of separating texture features from the substation monitoring image to be processed includes: Obtain the R channel value, G channel value, and B channel value of each pixel in the monitoring image of the substation to be processed; The R channel value, G channel value, and B channel value are converted into luminance channel value and chrominance channel value; Texture features are obtained by performing texture separation on the substation monitoring image to be processed based on the luminance channel value and color difference channel value corresponding to all pixels.

3. The method according to claim 1, characterized in that, The method further includes: Construct a filter, the filter including a conversion module, a high-frequency filtering module and a low-frequency filtering module, the conversion module being cascaded with the high-frequency filtering module and the low-frequency filtering module respectively; The step of converting the texture features to the frequency domain space to determine high-frequency component clusters and low-frequency component clusters in the frequency domain space includes: The texture features are input into the conversion module, and the texture features are converted into frequency domain signals and input into the high-frequency filtering module and the low-frequency filtering module; The high-frequency filtering module extracts all high-frequency components from the frequency domain signal to form the high-frequency component cluster; The low-frequency filtering module extracts all low-frequency components from the frequency domain signal to form the low-frequency component cluster.

4. The method according to claim 3, characterized in that, The high-frequency filtering module extracts all high-frequency components from the frequency domain signal to form the high-frequency component cluster, including: The high-frequency filtering module divides the frequency domain signal into spatial frequency distributions and calculates the entropy of each spatial frequency distribution region to extract the frequency domain signal covered by the spatial frequency distribution region when the entropy is greater than a set entropy threshold, so as to form the high-frequency component cluster.

5. The method according to claim 3, characterized in that, The low-frequency filtering module extracts all low-frequency components from the frequency domain signal to form the low-frequency component cluster, including: The low-frequency filtering module divides the frequency domain signal into spatial frequency distributions and calculates the entropy of each spatial frequency distribution region to extract the frequency domain signal covered by the spatial frequency distribution region when the entropy is less than a set entropy threshold, so as to form the low-frequency component cluster.

6. The method according to any one of claims 1-5, characterized in that, The step of performing sparse coding on the high-frequency component cluster and the low-frequency component cluster respectively to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix includes: The high-frequency component cluster and the low-frequency component cluster are vectorized respectively to obtain the corresponding high-frequency component vector and low-frequency component vector. Based on the linear combination vector of the set basis functions, the high-frequency component vector and the low-frequency component vector are sparsely encoded respectively to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix.

7. The method according to claim 6, characterized in that, The step of projecting the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space to obtain the corresponding high-frequency rank vector distribution and low-frequency rank vector distribution includes: Projecting the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space yields the corresponding high-frequency rank tensor and low-frequency rank tensor. Based on the high-frequency rank tensor and the low-frequency rank tensor, statistically analyze the distribution of the high-frequency rank vector and the distribution of the low-frequency rank vector.

8. A noise reduction device for substation monitoring images, characterized in that, include: The texture separation unit is used to acquire the substation monitoring image to be processed and perform texture separation on the substation monitoring image to obtain texture features; A texture conversion unit is used to convert the texture features to a frequency domain space to determine high-frequency component clusters and low-frequency component clusters in the frequency domain space. A sparse coding unit is used to perform sparse coding on the high-frequency component cluster and the low-frequency component cluster respectively to obtain a high-frequency sparse coding array and a low-frequency sparse coding matrix. The projection unit is used to project the high-frequency sparse coding array and the low-frequency sparse coding matrix into the rank space to obtain the corresponding high-frequency rank vector distribution and low-frequency rank vector distribution. The noise removal unit is used to determine and remove noise in the substation monitoring image to be processed based on the high-frequency rank vector distribution and the low-frequency rank vector distribution, including: projecting the high-frequency rank vector distribution and the low-frequency rank vector distribution into a Gaussian space to obtain the corresponding Gaussian distribution; The maximum likelihood estimates of the Gaussian distribution corresponding to the high-frequency rank vector distribution and the low-frequency rank vector distribution are determined and compared with the set maximum likelihood threshold to identify and remove noise in the substation monitoring image to be processed.

9. An electronic device, characterized in that, include: One or more processors; Computer-readable media, configured to store one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors perform the method as described in any one of claims 1-7.

Images