Lightweight transmission line ultraviolet image efficient denoising method and system
By combining adaptive spatial domain filtering and lightweight deep convolutional neural networks, the denoising problem of ultraviolet images of power transmission lines under strong noise backgrounds is solved, achieving efficient and lightweight image processing and improving the accuracy and real-time application capability of corona discharge detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANCHANG INST OF TECH
- Filing Date
- 2026-02-27
- Publication Date
- 2026-06-05
Smart Images
Figure CN122155987A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of ultraviolet imaging technology for power transmission lines, and particularly relates to a lightweight method and system for efficient noise reduction of ultraviolet images of power transmission lines. Background Technology
[0002] Ultraviolet (UV) imaging technology is an important means of detecting defects such as corona discharge and partial discharge on the surface of insulators in high-voltage transmission lines. By capturing the solar-blind UV photons (240-280nm) emitted during corona discharge, these photons can be converted into visible light spot images. However, in actual outdoor inspections, the acquired UV images are highly susceptible to interference from complex environments, mainly including: 1) scattering of residual UV components in sunlight; 2) inherent dark current noise, thermal noise, and photon shot noise of the camera sensor; and 3) background noise enhancement caused by weather factors such as rain, fog, and haze. These noises often have similar brightness and spatial distribution to the light spots generated by actual discharges in the image, resulting in a significant reduction in the image signal-to-noise ratio (SNR). This poses great difficulties for subsequent automated discharge detection, localization, and quantification based on image analysis, and significantly increases the false alarm rate and false negative rate.
[0003] Traditional ultraviolet image denoising methods are mainly divided into two categories: spatial domain filtering and transform domain filtering. Spatial domain methods, such as mean filtering, median filtering, and Gaussian filtering, are computationally simple, but they tend to blur the edges of real light spots and lose details. Furthermore, their fixed parameters cannot adapt to the noise levels in different regions of the image. Transform domain methods (such as wavelet thresholding and nonlocal mean filtering) can achieve better results, but when processing ultraviolet images with complex noise characteristics and requiring precise preservation of small, weak-intensity light spots, a trade-off between denoising and detail preservation still exists. In addition, while deep learning-based denoising methods (such as DnCNN and U-Net) offer superior performance, their models typically have a large number of parameters and high computational complexity, making them difficult to deploy in real-time on mobile or embedded ultraviolet inspection devices with limited computing resources.
[0004] Therefore, existing technologies lack a denoising method specifically designed for the characteristics of ultraviolet images of transmission lines, capable of accurately distinguishing and preserving weak corona signals against a strong noise background, while also meeting the requirements of lightweight and high-efficiency processing. This has become a key bottleneck restricting the in-depth application of ultraviolet imaging technology in intelligent line inspection. Summary of the Invention
[0005] This invention aims to overcome the aforementioned shortcomings of existing technologies and provide a lightweight, efficient denoising method and system for ultraviolet images of power transmission lines. This method achieves maximum preservation of potential discharge spot signals while effectively suppressing complex background noise through hierarchical analysis and collaborative processing of noise characteristics and target signals. Furthermore, the entire processing flow is computationally efficient and suitable for real-time or near-real-time applications.
[0006] In a first aspect, the present invention provides a lightweight ultraviolet image denoising method for transmission lines, comprising: The original ultraviolet image of the target transmission line is acquired, and the original ultraviolet image is preprocessed to extract a first feature information containing the distribution of background environmental noise and a second feature information containing the suspected corona discharge spot area. Based on the first feature information, a spatial domain filtering operator adapted to the current environmental noise level is constructed to perform initial denoising on the original ultraviolet image to obtain a first intermediate image. Based on the second feature information, multiple candidate target regions are located from the first intermediate image; A lightweight deep convolutional neural network model is constructed. The deep convolutional neural network model adopts an encoder-decoder structure. The encoder part uses depthwise separable convolution and multi-scale pooling layers to extract multi-level features. The decoder part fuses the multi-level features through skip connections and performs upsampling to output a noise weight map with the same size as the input image. The first intermediate image and the corresponding candidate target region mask are input into the trained lightweight deep convolutional neural network model to obtain a fine noise weight map. Based on the refined noise weight map, the original ultraviolet image is reconstructed by weighting to suppress background noise while preserving the true structure and intensity information of the candidate target region to the greatest extent, and outputting the final denoised image.
[0007] Secondly, the present invention provides a lightweight ultraviolet image high-efficiency noise reduction system for transmission lines, comprising: The acquisition module is configured to acquire the original ultraviolet image of the target transmission line and preprocess the original ultraviolet image to extract a first feature information containing the distribution of background environmental noise and a second feature information containing the suspected corona discharge spot area. The denoising module is configured to construct a spatial domain filtering operator that adapts to the current environmental noise level based on the first feature information, and perform initial denoising on the original ultraviolet image to obtain a first intermediate image. The localization module is configured to locate multiple candidate target regions from the first intermediate image based on the second feature information; The building module is configured to build a lightweight deep convolutional neural network model. The deep convolutional neural network model adopts an encoder-decoder structure, wherein the encoder part uses depthwise separable convolution and multi-scale pooling layers to extract multi-level features, and the decoder part fuses the multi-level features through skip connections and upsamples them to output a noise weight map with the same size as the input image. The output module is configured to input the first intermediate image and the corresponding candidate target region mask into the trained lightweight deep convolutional neural network model to obtain a fine noise weight map; The reconstruction module is configured to perform weighted reconstruction on the original ultraviolet image based on the fine noise weight map, so as to suppress background environmental noise while preserving the true structure and intensity information of the candidate target region to the greatest extent, and output the final denoised image.
[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the lightweight transmission line ultraviolet image high-efficiency denoising method according to any embodiment of the present invention.
[0009] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the lightweight transmission line ultraviolet image high-efficiency denoising method according to any embodiment of the present invention.
[0010] This application presents a lightweight method and system for efficient denoising of ultraviolet images of transmission lines. It utilizes extracted environmental noise features for adaptive spatial filtering, achieving rapid and targeted suppression of large-area background noise, laying a solid foundation for subsequent processing. Secondly, by locating candidate target regions and guiding a lightweight deep neural network for refined evaluation, the model accurately learns the differences between noise and real corona signals in local structure and context, thereby generating a weight map reflecting pixel-level signal-to-noise ratios. Finally, based on this weight map, adaptive weighted fusion is performed on the original image and the smoothed image, thoroughly suppressing background noise while preserving the intensity and morphological integrity of weak discharge spots to the greatest extent, significantly improving image quality. Compared to traditional methods, this invention achieves the best balance between detail preservation and noise suppression in denoising. The lightweight design, such as depthwise separable convolution, significantly reduces the number of model parameters and computational complexity, enabling real-time processing on mobile inspection terminals or embedded devices, meeting the practical needs of field operations. This method fundamentally improves the accuracy and reliability of subsequent automatic corona discharge identification and quantitative analysis, providing high-quality data preprocessing support for intelligent inspection of transmission lines. Attached Figure Description
[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0012] Figure 1 A flowchart of a lightweight transmission line ultraviolet image high-efficiency noise reduction method provided in an embodiment of the present invention; Figure 2 This is a structural block diagram of a lightweight ultraviolet image high-efficiency noise reduction system for transmission lines, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0013] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0014] Please see Figure 1 The flowchart illustrates a lightweight ultraviolet image denoising method for transmission lines according to this application.
[0015] like Figure 1 As shown, the efficient denoising method for ultraviolet images of lightweight transmission lines specifically includes the following steps: Step S101: Obtain the original ultraviolet image of the target transmission line and preprocess the original ultraviolet image to extract first feature information containing the distribution of background environmental noise and second feature information containing the suspected corona discharge spot area.
[0016] In this step, the local grayscale standard deviation and gradient magnitude of the original ultraviolet image at multiple scales are calculated to generate an environmental noise intensity distribution map, which is used as the first feature information. A segmentation algorithm based on local contrast enhancement and adaptive dual thresholds is adopted to initially extract connected regions in the image with significantly higher intensity than the local background. The area, mean intensity, and morphological irregularity of each region are calculated as the second feature information.
[0017] In one specific embodiment, a raw ultraviolet image I_original of the transmission line is acquired by an ultraviolet imager. Preprocessing includes: Extract the first feature information (environmental noise feature): Calculate the local gray - level standard deviation and gradient magnitude matrix of the image under sliding windows at different scales. The local gray - level standard deviation reflects the intensity of regional uniform noise, and the gradient magnitude can also reflect noise fluctuations in non - edge regions. Combine the two to generate an "environmental noise intensity distribution map" F_noise, which quantifies the background noise level at each position of the image.
[0018] Extract the second feature information (suspected target feature): Adopt a segmentation algorithm based on local contrast enhancement (such as CLAHE) and adaptive double - threshold (such as based on local mean and standard deviation) to preliminarily extract all connected regions in the image whose brightness is significantly higher than the surrounding background. Calculate the area Area, average gray - level Intensity_mean, and morphological irregularity (such as the ratio of area to the square of the perimeter) of each connected region. These regional information constitute the initial candidate target set Candidates.
[0019] Step S102, based on the first feature information, construct a spatial - domain filtering operator adaptive to the current environmental noise level, and perform primary denoising on the original ultraviolet image to obtain a first intermediate image.
[0020] In this step, according to the environmental noise intensity distribution map, divide the image into high - noise regions, medium - noise regions, and low - noise regions; Configure a first - type filtering kernel for the high - noise region, where the first - type filtering kernel has a larger size and a stronger smoothing coefficient; configure a second - type filtering kernel for the medium - noise region, where the second - type filtering kernel has a medium size and a smoothing coefficient; configure a third - type filtering kernel for the low - noise region, where the third - type filtering kernel has a smaller size or adopts edge - preserving characteristics; Based on the divided regions and the configured filtering kernels, perform non - uniform filtering processing on the original ultraviolet image pixel - by - pixel or in blocks to obtain the first intermediate image.
[0021] In a specific embodiment, according to the numerical distribution of F_noise, set two thresholds T_high and T_low, and divide the image pixels into three regions: high - noise region R_high (F_noise > T_high), medium - noise region R_mid (T_low < F_noise <= T_high), and low - noise region R_low (F_noise <= T_low).
[0022] Operator configuration and filtering: Configure a large Gaussian or mean filter kernel (e.g., 7x7) for R_high to perform strong smoothing; configure a medium-sized filter kernel (e.g., 5x5) for R_mid; for R_low, to preserve any possible details, a smaller filter kernel (e.g., 3x3) can be used, or the original pixel values can be used directly (i.e., no filtering). Based on the partitioning results, perform non-uniform filtering on I_original to obtain the first intermediate image I_smoothed. This step can effectively suppress large areas of uniform or gradually changing noise.
[0023] Step S103: Based on the second feature information, locate multiple candidate target regions from the first intermediate image.
[0024] In this step, the second feature information Candidates extracted in step S101 is used to further refine the image I_smoothed after initial denoising. For example, area and intensity thresholds are set to remove regions in Candidates that are too small in area or have too low average intensity (these are likely residual noise blocks). The remaining regions are marked as high-confidence candidate target regions, and a binary mask image Mask_target is generated, where the target region pixel value is 1 and the background pixel value is 0.
[0025] Step S104: Construct a lightweight deep convolutional neural network model. The deep convolutional neural network model adopts an encoder-decoder structure. The encoder part uses depthwise separable convolution and multi-scale pooling layers to extract multi-level features. The decoder part fuses the multi-level features through skip connections and performs upsampling to output a noise weight map with the same size as the input image.
[0026] In this step, the encoder part of the lightweight deep convolutional neural network model specifically includes: an input layer, a first depthwise separable convolutional module, a first multi-scale pooling layer, a second depthwise separable convolutional module, a second multi-scale pooling layer, and a bottleneck layer depthwise separable convolutional module connected in sequence. The multi-scale pooling layer performs max pooling and average pooling operations in parallel, and then outputs the results by concatenating them.
[0027] The decoder part of the lightweight deep convolutional neural network model specifically includes: a first upsampling layer, a first feature fusion layer, a second upsampling layer, a second feature fusion layer, and an output convolutional layer connected in sequence; The first feature fusion layer is used to fuse the skip connection features from the second multi-scale pooling layer with the features from the first upsampling layer; the second feature fusion layer is used to fuse the skip connection features from the first multi-scale pooling layer with the features from the second upsampling layer.
[0028] In one specific embodiment, before constructing a lightweight deep convolutional neural network model, a training sample set is constructed, wherein each sample includes: a paired noisy ultraviolet image, a first intermediate image and a candidate target region mask, and a corresponding real noise weight map label, wherein the real noise weight map label is obtained by calculating the pixel-level difference between the noisy image and the clean reference image and combining it with the discharge region labeled by experts. Using the training sample set, the lightweight deep convolutional neural network model is trained end-to-end with the goal of minimizing the loss function between the noise weight map predicted by the model and the labels of the real noise weight map.
[0029] Step S105: Input the first intermediate image and the corresponding candidate target region mask into the trained lightweight deep convolutional neural network model to obtain a fine noise weight map.
[0030] Step S106: Based on the fine noise weight map, the original ultraviolet image is reconstructed by weighting to suppress background noise while preserving the true structure and intensity information of the candidate target region to the greatest extent, and the final denoised image is output.
[0031] In this step, the formula for weighted reconstruction of the original ultraviolet image based on the fine noise weight map is as follows: I_out = I_original * W + I_smoothed * (1 - W) Where I_out is the final denoised image, I_original is the original ultraviolet image, I_smoothed is the first intermediate image, and W is the fine noise weight map, with pixel values between 0 and 1.
[0032] In summary, the method of this application utilizes extracted environmental noise features for adaptive spatial filtering, achieving rapid and targeted suppression of large-area background noise, laying a solid foundation for subsequent processing. Secondly, by locating candidate target regions and guiding a lightweight deep neural network for refined evaluation, the model can accurately learn the differences between noise and real corona signals in local structure and context, thereby generating a weight map reflecting pixel-level signal-to-noise probability. Finally, based on this weight map, adaptive weighted fusion of the original image and the smoothed image is performed, thoroughly suppressing background noise while preserving the intensity and morphological integrity of weak discharge spots to the greatest extent, significantly improving image quality. Compared to traditional methods, this invention achieves the best balance between detail preservation and noise suppression in denoising; the lightweight design, such as depthwise separable convolution, significantly reduces the number of model parameters and computational complexity, enabling real-time processing on mobile inspection terminals or embedded devices, meeting the practical needs of field operations. This method fundamentally improves the accuracy and reliability of subsequent automatic corona discharge identification and quantitative analysis, providing high-quality data preprocessing support for intelligent inspection of transmission lines.
[0033] Please see Figure 2 The diagram shows a structural block diagram of a lightweight ultraviolet image high-efficiency denoising system for transmission lines according to this application.
[0034] like Figure 2 As shown, the lightweight transmission line ultraviolet image high-efficiency denoising system 200 includes an acquisition module 210, a denoising module 220, a positioning module 230, a construction module 240, an output module 250, and a reconstruction module 260.
[0035] The acquisition module 210 is configured to acquire the original ultraviolet image of the target transmission line and preprocess the original ultraviolet image to extract first feature information containing background environmental noise distribution and second feature information containing suspected corona discharge spot regions; the denoising module 220 is configured to construct a spatial domain filtering operator adapted to the current environmental noise level based on the first feature information to perform initial denoising on the original ultraviolet image to obtain a first intermediate image; the localization module 230 is configured to locate multiple candidate target regions from the first intermediate image based on the second feature information; and the construction module 240 is configured to construct a lightweight deep convolutional neural network model. An encoder-decoder structure is adopted, wherein the encoder part uses depthwise separable convolution and multi-scale pooling layers to extract multi-level features, and the decoder part fuses the multi-level features through skip connections and performs upsampling to output a noise weight map with the same size as the input image; the output module 250 is configured to input the first intermediate image and the corresponding candidate target region mask into the trained lightweight deep convolutional neural network model to obtain a fine noise weight map; the reconstruction module 260 is configured to perform weighted reconstruction on the original ultraviolet image based on the fine noise weight map, so as to suppress background environmental noise while preserving the true structure and intensity information of the candidate target region to the greatest extent, and output the final denoised image.
[0036] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.
[0037] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the lightweight transmission line ultraviolet image high-efficiency noise reduction method in any of the above method embodiments. In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows: The original ultraviolet image of the target transmission line is acquired, and the original ultraviolet image is preprocessed to extract a first feature information containing the distribution of background environmental noise and a second feature information containing the suspected corona discharge spot area. Based on the first feature information, a spatial domain filtering operator adapted to the current environmental noise level is constructed to perform initial denoising on the original ultraviolet image to obtain a first intermediate image. Based on the second feature information, multiple candidate target regions are located from the first intermediate image; A lightweight deep convolutional neural network model is constructed. The deep convolutional neural network model adopts an encoder-decoder structure. The encoder part uses depthwise separable convolution and multi-scale pooling layers to extract multi-level features. The decoder part fuses the multi-level features through skip connections and performs upsampling to output a noise weight map with the same size as the input image. The first intermediate image and the corresponding candidate target region mask are input into the trained lightweight deep convolutional neural network model to obtain a fine noise weight map. Based on the refined noise weight map, the original ultraviolet image is reconstructed by weighting to suppress background noise while preserving the true structure and intensity information of the candidate target region to the greatest extent, and outputting the final denoised image.
[0038] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the lightweight transmission line ultraviolet image high-efficiency denoising system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely configured relative to a processor, which can be connected to the lightweight transmission line ultraviolet image high-efficiency denoising system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
[0039] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the lightweight transmission line ultraviolet image high-efficiency denoising method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the lightweight transmission line ultraviolet image high-efficiency denoising system. The output device 340 may include a display screen or other display device.
[0040] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.
[0041] In one implementation, the above-described electronic device is applied in a lightweight transmission line ultraviolet image high-efficiency noise reduction system for a client, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: The original ultraviolet image of the target transmission line is acquired, and the original ultraviolet image is preprocessed to extract a first feature information containing the distribution of background environmental noise and a second feature information containing the suspected corona discharge spot area. Based on the first feature information, a spatial domain filtering operator adapted to the current environmental noise level is constructed to perform initial denoising on the original ultraviolet image to obtain a first intermediate image. Based on the second feature information, multiple candidate target regions are located from the first intermediate image; A lightweight deep convolutional neural network model is constructed. The deep convolutional neural network model adopts an encoder-decoder structure. The encoder part uses depthwise separable convolution and multi-scale pooling layers to extract multi-level features. The decoder part fuses the multi-level features through skip connections and performs upsampling to output a noise weight map with the same size as the input image. The first intermediate image and the corresponding candidate target region mask are input into the trained lightweight deep convolutional neural network model to obtain a fine noise weight map. Based on the refined noise weight map, the original ultraviolet image is reconstructed by weighting to suppress background noise while preserving the true structure and intensity information of the candidate target region to the greatest extent, and outputting the final denoised image.
[0042] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.
[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A lightweight ultraviolet image denoising method for transmission lines, characterized in that, include: The original ultraviolet image of the target transmission line is acquired, and the original ultraviolet image is preprocessed to extract a first feature information containing the distribution of background environmental noise and a second feature information containing the suspected corona discharge spot area. Based on the first feature information, a spatial domain filtering operator adapted to the current environmental noise level is constructed to perform initial denoising on the original ultraviolet image to obtain a first intermediate image. Based on the second feature information, multiple candidate target regions are located from the first intermediate image; A lightweight deep convolutional neural network model is constructed. The deep convolutional neural network model adopts an encoder-decoder structure. The encoder part uses depthwise separable convolution and multi-scale pooling layers to extract multi-level features. The decoder part fuses the multi-level features through skip connections and performs upsampling to output a noise weight map with the same size as the input image. The first intermediate image and the corresponding candidate target region mask are input into the trained lightweight deep convolutional neural network model to obtain a fine noise weight map. Based on the refined noise weight map, the original ultraviolet image is reconstructed by weighting to suppress background noise while preserving the true structure and intensity information of the candidate target region to the greatest extent, and outputting the final denoised image.
2. The efficient noise reduction method for ultraviolet images of lightweight transmission lines according to claim 1, characterized in that, The preprocessing of the original ultraviolet image to extract the first feature information and the second feature information includes: Calculate the local grayscale standard deviation and gradient magnitude of the original ultraviolet image at multiple scales to generate an environmental noise intensity distribution map, which is used as the first feature information; A segmentation algorithm based on local contrast enhancement and adaptive dual thresholds is adopted to initially extract connected regions in the image with significantly higher intensity than the local background. The area, mean intensity, and morphological irregularity of each region are calculated as the second feature information.
3. The efficient noise reduction method for ultraviolet images of lightweight transmission lines according to claim 1, characterized in that, The step of constructing a spatial domain filtering operator adapted to the current environmental noise level based on the first feature information to perform initial denoising on the original ultraviolet image to obtain a first intermediate image includes: Based on the environmental noise intensity distribution map, the image is divided into high noise area, medium noise area and low noise area; A first type of filter core is configured for the high-noise region, which has a large size and a strong smoothing coefficient; a second type of filter core is configured for the medium-noise region, which has a medium size and a smoothing coefficient; and a third type of filter core is configured for the low-noise region, which has a small size or uses edge-preserving characteristics. Based on the divided regions and configured filter kernels, the original ultraviolet image is subjected to non-uniform filtering processing on a pixel-by-pixel or block-by-block basis to obtain the first intermediate image.
4. The efficient noise reduction method for ultraviolet images of lightweight transmission lines according to claim 1, characterized in that, The encoder portion of the lightweight deep convolutional neural network model specifically includes: The input layer, the first depthwise separable convolutional module, the first multi-scale pooling layer, the second depthwise separable convolutional module, the second multi-scale pooling layer, and the bottleneck layer depthwise separable convolutional module are connected in sequence. The multi-scale pooling layer performs max pooling and average pooling operations in parallel, and then outputs the results by concatenating them.
5. The efficient noise reduction method for ultraviolet images of lightweight transmission lines according to claim 1, characterized in that, The decoder portion of the lightweight deep convolutional neural network model specifically includes: The first upsampling layer, the first feature fusion layer, the second upsampling layer, the second feature fusion layer, and the output convolutional layer are connected in sequence. The first feature fusion layer is used to fuse the skip connection features from the second multi-scale pooling layer with the features from the first upsampling layer; the second feature fusion layer is used to fuse the skip connection features from the first multi-scale pooling layer with the features from the second upsampling layer.
6. The efficient noise reduction method for ultraviolet images of lightweight transmission lines according to claim 1, characterized in that, Before constructing a lightweight deep convolutional neural network model, the method further includes: A training sample set is constructed, wherein each sample includes: a paired noisy ultraviolet image, a first intermediate image and a candidate target region mask, and a corresponding real noise weight map label. The real noise weight map label is obtained by calculating the pixel-level difference between the noisy image and the clean reference image and combining it with the discharge region labeled by experts. Using the training sample set, the lightweight deep convolutional neural network model is trained end-to-end with the goal of minimizing the loss function between the noise weight map predicted by the model and the labels of the real noise weight map.
7. The efficient noise reduction method for ultraviolet images of lightweight transmission lines according to claim 1, characterized in that, The calculation formula for weighted reconstruction of the original ultraviolet image based on the fine noise weight map is as follows: I_out = I_original * W + I_smoothed * (1 - W) Where I_out is the final denoised image, I_original is the original ultraviolet image, I_smoothed is the first intermediate image, and W is the fine noise weight map, with pixel values between 0 and 1.
8. A lightweight ultraviolet image high-efficiency noise reduction system for transmission lines, characterized in that, include: The acquisition module is configured to acquire the original ultraviolet image of the target transmission line and preprocess the original ultraviolet image to extract a first feature information containing the distribution of background environmental noise and a second feature information containing the suspected corona discharge spot area. The denoising module is configured to construct a spatial domain filtering operator that adapts to the current environmental noise level based on the first feature information, and perform initial denoising on the original ultraviolet image to obtain a first intermediate image. The localization module is configured to locate multiple candidate target regions from the first intermediate image based on the second feature information; The building module is configured to build a lightweight deep convolutional neural network model. The deep convolutional neural network model adopts an encoder-decoder structure, wherein the encoder part uses depthwise separable convolution and multi-scale pooling layers to extract multi-level features, and the decoder part fuses the multi-level features through skip connections and upsamples them to output a noise weight map with the same size as the input image. The output module is configured to input the first intermediate image and the corresponding candidate target region mask into the trained lightweight deep convolutional neural network model to obtain a fine noise weight map; The reconstruction module is configured to perform weighted reconstruction on the original ultraviolet image based on the fine noise weight map, so as to suppress background environmental noise while preserving the true structure and intensity information of the candidate target region to the greatest extent, and output the final denoised image.
9. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method according to any one of claims 1 to 7.