Digital compression method for power monitoring image

Abstract

The invention discloses a digital compression method for a power monitoring image, and the method comprises the following steps: image preprocessing: carrying out the preprocessing of an input power monitoring image, including denoising and contrast enhancement, and obtaining a preprocessed power monitoring image; feature extraction: extracting key features of the preprocessed power monitoring image by using a deep learning model; feature compression: compressing the key features extracted in the step S2 to obtain compressed power monitoring image features; feature reconstruction: decompressing the compressed power monitoring image, and reconstructing the image by using the decompressed power monitoring image features; and post-processing: carrying out post-processing on the reconstructed image, wherein the post-processing comprises removing artifacts and enhancing details. According to the method, efficient compression and high-quality reconstruction of the power monitoring image are realized through deep learning and image processing technologies.

Description

Technical Field

[0001] The invention relates to the field of power image processing, in particular to a digital compression method for power monitoring images. Background Art

[0002] In the field of power monitoring, efficient storage and transmission of image data is an important research topic. Power monitoring images usually contain a lot of detailed information, such as the appearance of power equipment, the connection status of lines, etc., which is crucial for the safe operation of the power system. However, due to the high resolution and complexity of power monitoring images, directly storing and transmitting these images will take up a lot of storage space and bandwidth resources, resulting in inefficient data processing.

[0003] Traditional image compression methods, such as JPEG and JPEG2000, have achieved good results in the field of general image compression, but they have some limitations when processing power monitoring images. First, these methods usually rely on the general features of images and cannot fully utilize the particularity of power monitoring images, resulting in low compression efficiency. Second, traditional compression methods often introduce certain distortions during the compression process, which is not conducive to the accurate analysis and diagnosis of power monitoring images. Summary of the invention

[0004] The purpose of the invention is to provide a digital compression method for power monitoring images to solve the above-mentioned problems existing in the prior art.

[0005] Technical solution: A digital compression method for power monitoring images, comprising the following steps:

[0006] Step S1, image preprocessing: preprocessing the input power monitoring image, including denoising and contrast enhancement, to obtain a preprocessed power monitoring image;

[0007] Step S2, feature extraction: extracting key features of the preprocessed power monitoring image using a deep learning model;

[0008] Step S3, feature compression: compressing the key features extracted in step S2 to obtain compressed power monitoring image features;

[0009] Step S4, feature reconstruction: decompressing the compressed power monitoring image, and reconstructing the image using the features of the decompressed power monitoring image;

[0010] Step S5, post-processing: post-processing the reconstructed image, including removing artifacts and enhancing details.

[0011] According to another aspect of the present application, the step S1 is further:

[0012] Step S11, image format conversion: converting the input power monitoring image into a unified format;

[0013] Step S12, image resolution adjustment: adjusting the image to a resolution suitable for subsequent processing;

[0014] Step S13, image denoising: Use median filtering to remove noise in the image. The specific method is: for each pixel point in the image, take the median value in its neighborhood as the value of the point. The formula is as follows:

[0015] I ′ (x,y)=median{I(x+i,y+j)|i,j∈[-k,k]};

[0016] Among them, median is the median filter, k is the size of the neighborhood, (x, y) is the pixel point, and i, j are integers belonging to the neighborhood;

[0017] Step S14, contrast enhancement: using adaptive histogram equalization to enhance the image contrast, the specific method is: performing histogram equalization on each local area of ​​the image, the formula is as follows:

[0018] I ′ (x,y)=CDFlocal(I(x,y))×(L-1);

[0019] Among them, CDFlocal is the cumulative distribution function of the local area, and L is the number of gray levels.

[0020] According to another aspect of the present application, step S2 is further:

[0021] Step S21, input layer: input pre-processed power monitoring image;

[0022] Step S22, convolutional layer: Use multiple convolutional layers to extract features at different levels. The formula is as follows:

[0023] F l+1 (x,y)=σ(∑ i,j F l (x+i,y+j)·W l+1 (i,j)+b l+1 );

[0024] Among them, F l is the feature map of layer l, W l+1 is the convolution kernel, b l+1 is the bias, σ is the activation function;

[0025] Step S23, pooling layer: Use maximum pooling to reduce the size of the feature map. The formula is as follows:

[0026] Fl+1 (x,y)=max{F l (x+i,y+j)|i,j∈[-k,k]};

[0027] Where k is the size of the pooling kernel;

[0028] Step S24, fully connected layer: flatten the feature map and connect it to the fully connected layer, output the feature vector, the formula is as follows:

[0029] F fc =σ(∑ i F conv (i) W fc (i)+b fc )

[0030] Among them, F conv is the output feature map of the convolutional layer, W fc is the weight of the fully connected layer, b fc is bias.

[0031] According to another aspect of the present application, step S3 is further:

[0032] Step S31, quantization: quantize the feature vector to reduce data accuracy. The formula is as follows:

[0033] F quant (i) = round(F fc (i) / Q)×Q;

[0034] Where Q is the quantization step size;

[0035] Step S32, entropy coding: use Huffman coding or arithmetic coding to perform lossless compression on the quantized features.

[0036] According to another aspect of the present application, step S4 is further:

[0037] Step S41, decompression: decompressing the compressed features to restore the original feature vectors;

[0038] Step S42, deconvolution: Use the deconvolution layer, i.e., transposed convolution, to reconstruct the feature vector into an image. The formula is as follows:

[0039] F l+1 (x,y)=σ(∑ i,j F l (xi,yj)·W l+1 (i,j)+b l+1 );

[0040] Among them, Fl is the feature map of the lth layer, W l+1 is the deconvolution kernel, bl+1 is the bias and σ is the activation function.

[0041] According to another aspect of the present application, the step S5 is further as follows: Step S51, removing artifacts: using a removal artifact algorithm to remove artifacts in the reconstructed image, the formula is as follows:

[0042] I ′ (x,y)=IDWT(DWT(I(x,y)));

[0043] Among them, DWT is wavelet transform, IDWT is inverse wavelet transform;

[0044] Step S52, detail enhancement: Use a sharpening filter to enhance image details. The formula is as follows:

[0045]

[0046] in, is the Laplacian operator and α is the sharpening strength.

[0047] According to another aspect of the present application, a digital compression system for power monitoring images is also provided, comprising:

[0048] at least one processor; and

[0049] a memory communicatively connected to at least one of the processors; wherein,

[0050] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement a digital compression method for power monitoring images described in any of the above technical solutions.

[0051] Beneficial effects: The present invention extracts key features of power monitoring images through a deep learning model, and combines entropy coding technology to achieve a higher compression ratio; the present invention realizes high-quality image reconstruction through a deep learning model and image processing technology, and the reconstructed image is superior to traditional methods in terms of detail retention and visual quality; the present invention automatically learns high-level features of images through a deep learning model, and can better adapt to power monitoring images in different scenes and lighting conditions, and has strong adaptability; the present invention improves real-time processing capabilities by optimizing deep learning models and compression algorithms, and can achieve efficient image compression and transmission in power monitoring systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is an overall flow chart of the solution of the present invention.

[0053] Figure 2 It is a flowchart of step S1 of the present invention.

[0054] Figure 3 This is a flow chart of step S2 of the present invention.

[0055] Figure 4 This is a flow chart of step S3 of the present invention.

[0056] Figure 5 This is a flow chart of step S4 of the present invention.

[0057] Figure 6 This is a flow chart of step S5 of the present invention. DETAILED DESCRIPTION

[0058] like Figure 1 As shown, according to one aspect of the present application, a method for digital compression of a power monitoring image comprises the following steps:

[0059] Step S1, image preprocessing: preprocessing the input power monitoring image, including denoising and contrast enhancement, to obtain a preprocessed power monitoring image;

[0060] Step S2, feature extraction: extracting key features of the preprocessed power monitoring image using a deep learning model;

[0061] Step S3, feature compression: compressing the key features extracted in step S2 to obtain compressed power monitoring image features;

[0062] Step S4, feature reconstruction: decompressing the compressed power monitoring image, and reconstructing the image using the features of the decompressed power monitoring image;

[0063] Step S5, post-processing: post-processing the reconstructed image, including removing artifacts and enhancing details.

[0064] Electricity monitoring images are usually high-resolution and complex. Direct storage and transmission of these images will take up a lot of storage space and bandwidth resources, resulting in low data processing efficiency. Traditional compression methods (such as JPEG, JPEG2000) often fail to fully utilize the particularity of power images when processing power monitoring images, resulting in low compression efficiency. This embodiment extracts the key features of power monitoring images through a deep learning model, and combines entropy coding technology to achieve a higher compression ratio, significantly reduce the amount of data in power monitoring images, and improve data storage and transmission efficiency.

[0065] The detailed information of power monitoring images is crucial to the safe operation of the power system. Traditional compression methods often introduce certain distortions during the compression process, resulting in a decrease in image quality, which is not conducive to the accurate analysis and diagnosis of power monitoring images. This embodiment achieves high-quality image reconstruction through deep learning models and image processing technology. The reconstructed image is superior to traditional methods in terms of detail retention and visual quality, effectively solving the problem of image quality loss.

[0066] The scenes and lighting conditions of power monitoring images are diverse, and the performance of traditional compression methods in different scenes may vary greatly, and their adaptability is poor. This embodiment automatically learns the advanced features of images through a deep learning model, which can better adapt to power monitoring images in different scenes and lighting conditions, has strong adaptability, and solves the problem of poor adaptability.

[0067] The power monitoring system usually needs to process a large amount of image data in real time, and the traditional compression method has certain limitations in real-time processing. This embodiment improves the real-time processing capability by optimizing the deep learning model and compression algorithm, and can achieve efficient image compression and transmission in the power monitoring system, solving the problem of insufficient real-time processing capability.

[0068] The stability of the power monitoring system is crucial to the safe operation of the power system. Traditional compression methods may introduce certain delays and distortions when processing large amounts of image data, affecting the stability of the system. This embodiment reduces the time and distortion of data processing through efficient compression and reconstruction methods, enhances the stability of the power monitoring system, and solves the problem of system stability.

[0069] Traditional compression methods usually rely on fixed algorithms and parameters, are difficult to integrate into existing power monitoring systems, and have poor scalability. The method of this embodiment is based on deep learning and image processing technology, has a good modular design, and is easy to integrate into existing power monitoring systems. At the same time, this method can also achieve functional expansion and performance improvement by further optimizing the deep learning model and compression algorithm, solving the problem of integration and expansion difficulties.

[0070] In summary, this embodiment proposes a new digital compression method for power monitoring images through deep learning and image processing technology, which solves a series of problems faced by power monitoring images during storage and transmission, such as low data storage and transmission efficiency, image quality loss, poor adaptability, insufficient real-time processing capabilities, system stability problems, and integration and expansion difficulties. Through these improvements, this embodiment provides a new solution for data management of power monitoring systems, which has important practical value.

[0071] According to another aspect of the present application, the step S1 is further:

[0072] Step S11, image format conversion: converting the input power monitoring image into a unified format;

[0073] Step S12, image resolution adjustment: adjusting the image to a resolution suitable for subsequent processing;

[0074] Step S13, image denoising: Use median filtering to remove noise in the image. The specific method is: for each pixel point in the image, take the median value in its neighborhood as the value of the point. The formula is as follows:

[0075] I ′ (x,y)=median{I(x+i,y+j)|i,j∈[-k,k]};

[0076] Among them, median is the median filter, k is the size of the neighborhood, (x, y) is the pixel point, and i, j are integers belonging to the neighborhood;

[0077] Step S14, contrast enhancement: using adaptive histogram equalization to enhance the image contrast, the specific method is: performing histogram equalization on each local area of ​​the image, the formula is as follows:

[0078] I ′ (x,y)=CDFlocal(I(x,y))×(L-1);

[0079] Among them, CDFlocal is the cumulative distribution function of the local area, and L is the number of gray levels.

[0080] This embodiment solves the problems of inconsistent formats, inconsistent resolutions, noise interference, and insufficient contrast faced by power monitoring images during processing through four sub-steps: image format conversion, image resolution adjustment, image denoising, and contrast enhancement. These improvements not only improve image quality and processing efficiency, but also ensure the consistency and stability of subsequent processing steps, providing a solid foundation for data management of power monitoring systems.

[0081] According to another aspect of the present application, step S2 is further:

[0082] Step S21, input layer: input pre-processed power monitoring image;

[0083] Step S22, convolutional layer: Use multiple convolutional layers to extract features at different levels. The formula is as follows:

[0084] F l+1 (x,y)=σ(∑ i,j F l (x+i,y+j)·W l+1 (i,j)+bl+1 );

[0085] Among them, F l is the feature map of layer l, W l+1 is the convolution kernel, b l+1 is the bias, σ is the activation function;

[0086] Step S23, pooling layer: Use maximum pooling to reduce the size of the feature map. The formula is as follows:

[0087] F l+1 (x,y)=max{F l (x+i,y+j)|i,j∈[-k,k]};

[0088] Where k is the size of the pooling kernel;

[0089] Step S24, fully connected layer: flatten the feature map and connect it to the fully connected layer, output the feature vector, the formula is as follows:

[0090] F fc =σ(∑ i F conv (i) W fc (i)+b fc )

[0091] Among them, F conv is the output feature map of the convolutional layer, W fc is the weight of the fully connected layer, b fc is bias.

[0092] This embodiment uses a deep learning model to extract key features of preprocessed power monitoring images, solving the problems of low feature extraction efficiency, limited feature expression ability, and poor adaptability faced by traditional feature extraction methods in power monitoring image processing. The deep learning model significantly improves the efficiency and accuracy of feature extraction through automatic learning of advanced features, end-to-end learning, and good scalability.

[0093] According to another aspect of the present application, step S3 is further:

[0094] Step S31, quantization: quantize the feature vector to reduce data accuracy. The formula is as follows:

[0095] F quant (i) = round(F fc (i) / Q)×Q;

[0096] Where Q is the quantization step size;

[0097] Step S32, entropy coding: use Huffman coding or arithmetic coding to perform lossless compression on the quantized features.

[0098] This embodiment achieves efficient compression of feature vectors through quantization and entropy coding technology, solving the problems of low compression efficiency, feature loss and poor adaptability faced by traditional feature compression methods in power monitoring image processing. Quantization technology maintains feature information by reducing data precision; entropy coding technology improves compression ratio and maintains data integrity through lossless compression. These improvements not only improve the efficiency and quality of feature compression, but also ensure the quality of the reconstructed image.

[0099] According to another aspect of the present application, step S4 is further:

[0100] Step S41, decompression: decompressing the compressed features to restore the original feature vectors;

[0101] Step S42, deconvolution: Use the deconvolution layer (transposed convolution) to reconstruct the feature vector into an image. The formula is as follows:

[0102] F l+1 (x,y)=σ(∑ i,j F l (xi,yj)·W l+1 (i,j)+b l+1 );

[0103] Among them, Fl is the feature map of the lth layer, W l+1 is the deconvolution kernel, b l+1 is the bias and σ is the activation function.

[0104] This embodiment achieves efficient reconstruction of feature vectors through deconvolution technology, solving the problems of low reconstruction efficiency, image quality loss and poor adaptability faced by traditional feature reconstruction methods in power monitoring image processing. Deconvolution technology significantly improves reconstruction efficiency and image quality through efficient reconstruction and high-quality reconstruction, ensuring that the details of the reconstructed image are preserved.

[0105] According to another aspect of the present application, step S5 is further:

[0106] Step S51, artifact removal: Use an artifact removal algorithm to remove artifacts in the reconstructed image. The formula is as follows:

[0107] I ′ (x,y)=IDWT(DWT(I(x,y)));

[0108] Among them, DWT is wavelet transform, IDWT is inverse wavelet transform;

[0109] Step S52, detail enhancement: Use a sharpening filter to enhance image details. The formula is as follows:

[0110]

[0111] in, is the Laplacian operator and α is the sharpening strength.

[0112] This embodiment achieves efficient post-processing of the reconstructed image through artifact removal and detail enhancement technology, solving the problems of poor artifact removal effect, insufficient detail enhancement and poor adaptability faced by traditional post-processing methods in power monitoring image processing. The artifact removal technology effectively removes artifacts from the reconstructed image through wavelet transform and maintains image details; the detail enhancement technology effectively enhances image details through sharpening filters, improving image clarity and detail visibility. These improvements not only improve the image quality and processing efficiency, but also ensure the visual effect of the reconstructed image.

[0113] According to another aspect of the present application, a digital compression system for power monitoring images is also provided, comprising:

[0114] at least one processor; and

[0115] a memory communicatively connected to at least one of the processors; wherein,

[0116] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement a digital compression method for power monitoring images described in any of the above technical solutions.

[0117] It should be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

Claims

1. A digital compression method for power monitoring images, characterized in that: The steps include: Step S1, image preprocessing: preprocessing the input power monitoring image, including denoising and contrast enhancement, to obtain a preprocessed power monitoring image; Step S2, feature extraction: extracting key features of the preprocessed power monitoring image using a deep learning model; Step S3, feature compression: compressing the key features extracted in step S2 to obtain compressed power monitoring image features; Step S4, feature reconstruction: decompressing the compressed power monitoring image, and reconstructing the image using the features of the decompressed power monitoring image; Step S5, post-processing: post-processing the reconstructed image, including removing artifacts and enhancing details.

2. A digital compression method for power monitoring images as claimed in claim 1, characterized in that: The image preprocessing step S1 is further as follows: Step S11, image format conversion: converting the input power monitoring image into a unified format; Step S12, image resolution adjustment: adjusting the image to a resolution suitable for subsequent processing; Step S13, image denoising: Use median filtering to remove noise in the image. The specific method is: for each pixel point in the image, take the median value in its neighborhood as the value of the point. The formula is as follows: I ′ (x,y)=median{I(x+i,y+j)∣i,j∈[-k,k]}; Among them, median is the median filter, k is the size of the neighborhood, (x, y) is the pixel point, and i, j are integers belonging to the neighborhood; Step S14, contrast enhancement: using adaptive histogram equalization to enhance the image contrast, the specific method is: performing histogram equalization on each local area of ​​the image, the formula is as follows: I ′ (x,y)=CDFlocal(I(x,y))×(L-1); Among them, CDFlocal is the cumulative distribution function of the local area, and L is the number of gray levels.

3. A digital compression method for power monitoring images as claimed in claim 2, characterized in that: The feature extraction method in step S2 is to design a multi-layer CNN model to extract high-level features of the image, including: Step S21, input layer: input pre-processed power monitoring image; Step S22, convolutional layer: Use multiple convolutional layers to extract features at different levels. The formula is as follows: F l+1 (x,y)=σ(∑ i,j F l (x+i,y+j)·W l+1 (i,j)+b l+1 ); Among them, F l is the feature map of layer l, W l+1 is the convolution kernel, b l+1 is the bias, σ is the activation function; Step S23, pooling layer: Use maximum pooling to reduce the size of the feature map. The formula is as follows: F l+1 (x,y)=max{F l (x+i,y+j)∣i,j∈[-k,k]}; Where k is the size of the pooling kernel; Step S24, fully connected layer: flatten the feature map and connect it to the fully connected layer, output the feature vector, the formula is as follows: F fc =σ(∑ i F conv (i)·W fc (i)+b fc ) Among them, F conv is the output feature map of the convolutional layer, W fc is the weight of the fully connected layer, b fc is bias.

4. A digital compression method for power monitoring images as claimed in claim 3, characterized in that: The feature compression in step S3 is further as follows: Step S31, quantization: quantize the feature vector to reduce data accuracy. The formula is as follows: F quant (i)=round(F fc (i) / Q)×Q; Where Q is the quantization step size; Step S32, entropy coding: use Huffman coding or arithmetic coding to perform lossless compression on the quantized features.

5. A digital compression method for power monitoring images as claimed in claim 4, characterized in that: The feature reconstruction in step S4 is further as follows: Step S41, decompression: decompressing the compressed features to restore the original feature vectors; Step S42, deconvolution: Use the deconvolution layer, i.e., transposed convolution, to reconstruct the feature vector into an image. The formula is as follows: F l+1 (x,y)=σ(∑ i,j F l (x-i,y-j)·W l+1 (i,j)+b l+1 ); Among them, Fl is the feature map of the lth layer, W l+1 is the deconvolution kernel, b l+1 is the bias and σ is the activation function.

6. A method for digital compression of electric power monitoring images as claimed in claim 5, characterized in that: The step S5 is further as follows: Step S51, artifact removal: Use an artifact removal algorithm to remove artifacts in the reconstructed image. The formula is as follows: I ′ (x,y)=IDWT(DWT(I(x,y))); Among them, DWT is wavelet transform, IDWT is inverse wavelet transform; Step S52, detail enhancement: Use a sharpening filter to enhance image details. The formula is as follows: in, is the Laplacian operator and α is the sharpening strength.

7. A digital compression system for power monitoring images, characterized in that: include: at least one processor; as well as a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the intelligent patrol method of substation equipment based on two-stage coordinated optimization as described in any one of claims 1 to 6.

Images