Image quality enhancement method and system

Abstract

The invention discloses an image quality enhancement method and system, and relates to the technical field of intelligent detection of a power system, and the method comprises the steps: inputting a preprocessed high-voltage line accident scene image into a multi-scale convolutional neural network, and obtaining a feature map; respectively inputting the feature map into a brightness branch, a contrast-texture branch and an electromagnetic-denoising branch to obtain a feature map of the corresponding branch; inputting the feature maps of all the branches into the trained gating network to obtain a fusion weight of each branch; and according to the fusion weight of each branch, performing weighted summation on the feature map of each branch to obtain an enhanced image. According to the invention, the quality of the high-voltage line accident scene image can be enhanced in a targeted manner.

Description

Technical Field

[0001] The present invention relates to the technical field of power system intelligent detection, and in particular to an image quality enhancement method and system. Background Art

[0002] With the rapid development of the power industry, high-voltage power line accidents are becoming more frequent, making rapid and accurate analysis of accident scenes crucial. Drones and unmanned robots, due to their flexibility and efficiency, are widely used to inspect and capture high-voltage power line incidents. However, accident scenes are complex environments, and images captured by drones and unmanned robots are often affected by factors such as insufficient lighting, smoke, and strong electromagnetic interference, resulting in poor image quality.

[0003] There are two existing technologies currently disclosed. Technology 1 is patent CN112241765A, which uses multi-scale convolution and attention mechanism to quickly and efficiently obtain classification results while obtaining high-precision verification results; Technology 2 is patent CN114863099A, which uses multi-branch asymmetric convolution for cloud image segmentation, and can achieve good segmentation effects when there is an imbalance in the data of daytime and nighttime cloud images. However, the above existing technologies do not perform adaptive image analysis for high-voltage scenes, resulting in poor quality of image processing results. Therefore, how to specifically enhance the quality of high-voltage power line accident scene images is an important issue that needs to be solved urgently. Summary of the Invention

[0004] The embodiment of the present invention provides an image quality enhancement method and system, which cannot specifically enhance the quality of images of high-voltage power line accident scenes.

[0005] An embodiment of the present invention provides a method for enhancing image quality, comprising the following steps: Obtain feature maps of high-voltage power line accident scene images; The first feature map is obtained by separating the arc glare and background in the feature map through an asymmetric mapping function. The second feature map is obtained by enhancing the image contrast of the feature map through the CLAHE algorithm and gradient guidance. The noise distribution in the feature map is learned through residual learning and denoised and smoke artifacts are suppressed using frequency domain masks to obtain the third feature map. The first, second and third feature maps are weighted and fused to obtain a quality enhanced image.

[0006] Furthermore, before obtaining the feature map of the high-voltage power line accident scene picture, the method further includes: pre-processing the captured high-voltage power line accident scene picture; The pre-processing step includes: using Fourier transform to convert the high-voltage power line accident scene picture from the time domain to the frequency domain, and filtering the frequency domain to remove noise; The filtering result is restored from the frequency domain to the time domain using inverse Fourier transform to obtain a filtered time domain image; The filtered time domain is input into the adaptive convolution, and the convolution kernel parameters are dynamically adjusted according to the noise intensity to suppress the residual noise.

[0007] Furthermore, the quality-enhanced image is obtained using an image enhancement network, and the image enhancement network includes: a dynamic multi-task encoder, configured to input a pre-processed high-voltage power line accident scene image into the dynamic multi-task encoder to obtain a feature map of the high-voltage power line accident scene image; The brightness branch, contrast-texture branch, and electromagnetic-denoising branch are used to extract features from the feature maps of the high-voltage power line accident scene image to obtain the first, second, and third feature maps respectively; The gating network is used to input the first, second and third feature maps into the trained gating network for fusion to obtain a quality enhanced image.

[0008] Furthermore, the step of inputting the pre-processed high-voltage power line accident scene picture into a dynamic multi-task encoder to obtain a feature map of the high-voltage power line accident scene picture specifically includes: Based on the pre-processed high-voltage power line accident scene images, a lightweight convolutional neural network is used to extract the illumination and noise distribution features. Dynamically adjust the channel attention and spatial attention weights to enhance the fault area; according to the channel attention, adjust the feature strength of each channel through multiplication operation; according to the spatial attention, highlight the fault area through element-level multiplication or weighted summation; Obtain feature maps of high-voltage power line accident scene images.

[0009] Furthermore, the trained gating network specifically includes the following training steps: Obtain images of high-voltage power line accident scenes under different accident types; The gated network is trained using images of high-voltage power line accident scenes under different accident types to automatically adjust the weights of the brightness branch, contrast-texture branch, and electromagnetic-denoising branch in the fusion process. According to the fusion weight of each branch, the feature map of each branch is weighted summed to obtain the optimal fusion strategy under different accident types.

[0010] An embodiment of the present invention provides an image quality enhancement system, comprising: An image acquisition module is used to obtain a feature map of a high-voltage power line accident scene; The branch extraction module is used to separate the arc glare and background in the feature map through an asymmetric mapping function to obtain the first feature map; enhance the image contrast of the feature map through the CLAHE algorithm and gradient guidance to obtain the second feature map; and learn the noise distribution in the feature map through residual learning and use frequency domain masking to denoise and suppress smoke artifacts to obtain the third feature map; The image enhancement module is used to assign weights to the first, second and third feature maps and fuse them to obtain a quality-enhanced image.

[0011] The embodiments of the present invention provide a method and system for enhancing image quality. Compared with the prior art, the methods and systems have the following advantages: The arc glare and background in the feature map are separated by an asymmetric mapping function to obtain the first feature map; the image contrast of the feature map is enhanced by the CLAHE algorithm and gradient-guided convolution to obtain the second feature map; the noise distribution in the feature map is learned by residual learning and denoised and smoke artifacts are suppressed by frequency domain masking to obtain the third feature map; weights are then assigned to the first, second and third feature maps and fused to obtain a quality-enhanced image; the quality-enhanced image fuses the three key feature maps together, and in the process of weight assignment and fusion, a picture of the accident scene with the most important features is obtained in a targeted manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A schematic diagram of the system architecture provided by an embodiment of the present invention; Figure 2 A flowchart of electromagnetic noise modeling and suppression provided by an embodiment of the present invention; Figure 3 A flow chart of a dynamic multi-tasking encoder provided by an embodiment of the present invention; Figure 4 This is a flow chart of a multi-branch collaborative encoder provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0013] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0014] See also Figure 1 , an embodiment of the present invention provides an image quality enhancement method, comprising the following steps: Step 1: Obtain the feature map of the high-voltage power line accident scene image.

[0015] Step 2: Separate the arc glare and background in the feature map through an asymmetric mapping function to obtain the first feature map; enhance the image contrast of the feature map through the CLAHE algorithm and gradient guidance to obtain the second feature map; residual learning of the noise distribution in the feature map and use frequency domain masking to denoise and suppress smoke artifacts to obtain the third feature map.

[0016] Step 3: Assign weights to the first, second, and third feature maps and fuse them to obtain a quality-enhanced image.

[0017] This invention describes a low-quality image enhancement method for high-voltage power line accident scenes in complex environments (such as strong electromagnetic interference, dynamic lighting, and smoke / flame noise). It is suitable for real-time processing of power accident images collected by drones, inspection robots, and other equipment, providing high-precision visual support for arc detection, wire fracture location, and emergency repairs. This invention proposes a system that integrates electromagnetic noise suppression, dynamic attention allocation, and multi-task collaborative enhancement to address key challenges in high-voltage accident image enhancement.

[0018] The technical solution of the present invention is explained in detail as follows: 1. Electromagnetic noise suppression front end: like Figure 2 As shown in the figure, it includes two parts: the frequency domain filtering module and the adaptive convolution kernel. The frequency domain filtering module detects periodic stripe noise through Fourier transform and generates a frequency domain mask for filtering. The adaptive convolution kernel dynamically adjusts the convolution kernel parameters according to the noise intensity to suppress residual noise.

[0019] 2. Dynamic multi-tasking encoder: like Figure 3 As shown in the figure, through this dynamic attention adjustment process, the system can adaptively optimize the image enhancement effect according to the image content and scene conditions, so that key areas and details are better presented. It mainly includes the following four modules: (1) Scene perception attention module: This module dynamically adjusts the channel and spatial attention weights according to the lighting / noise distribution of the input image (such as arc intensity and smoke concentration).

[0020] Prioritize enhancing faulty areas (such as breakpoints) for subsequent decoder use.

[0021] (2) Light / noise distribution feature extraction: A lightweight convolutional neural network is used to perform preliminary feature extraction on the input image. The multi-scale feature extraction layer includes parallel 1×1, 3×3, and 5×5 convolutions, focusing on key factors such as light intensity, arc intensity, and smoke concentration.

[0022] These features are compressed into vectors through pooling layers (such as global average pooling) for subsequent attention weight calculation.

[0023] (3) Dynamic attention weight calculation: Channel attention: Calculates attention weights for each channel (e.g., RGB channels). Use the Sigmoid activation function to perform a nonlinear transformation on the channel feature vectors to obtain per-channel weights. These weights reflect the importance of different channels in the enhancement process.

[0024] Spatial Attention: In the spatial dimension, convolution operations (such as 3x3 convolution) are used to generate a spatial attention map. This map determines which areas require more attention (such as fault areas) by calculating the feature response at each spatial location. Sigmoid activation function is also used for normalization.

[0025] (4) Attention application: The calculated channel attention and spatial attention weights are applied to the feature maps respectively. Channel attention adjusts the feature strength of each channel through multiplication operations, while spatial attention highlights key areas through element-wise multiplication or weighted summation.

[0026] 3. Multi-branch collaborative decoder: like Figure 4 As shown in the figure, through this dynamic allocation process, the gating network can adaptively optimize the fusion strategy of each branch according to the content and scene conditions of the input image, thereby improving the effect and robustness of image enhancement. It mainly includes the following parts: (1) The gated network receives input: The gating network receives the output feature maps from the brightness enhancement branch, contrast-texture branch, and electromagnetic-denoising branch.

[0027] (2) Learning the optimal fusion strategy: The gating network is trained to learn the optimal fusion strategy for different types of incidents (such as arc discharge and wire breakage). This means that it automatically adjusts the importance of each branch in the fusion process based on the specific content of the input image.

[0028] Brightness Enhancement: Introduces an asymmetric mapping function to separate arc glare from background light, preventing overexposure. Contrast-Texture: Integrates the CLAHE algorithm with gradient-guided convolution to enhance wire edges. Electromagnetic Denoising: Combines residual learning with frequency-domain masking to suppress residual electromagnetic noise and smoke artifacts.

[0029] The specific training and learning process is as follows: end-to-end scene perception training, the gated network gradually masters the strategy of dynamically adjusting branch weights according to the scene through multi-task supervised learning and curriculum training strategies.

[0030] Step 1: Dataset construction (scenario-driven multimodal annotation).

[0031] 1) Raw data: Collect images of power accident scenes, covering: Arc discharge: strong light flare, metal vapor cloud, wire melting mark; wire breakage: flying debris, arc track, insulator damage; equipment short circuit: smoke cover, electromagnetic interference, local overheating traces.

[0032] 2) Annotation System: Professional Enhancement Images: Professionals annotate the ideal enhancement results, including a three-dimensional rating system for brightness balance, texture clarity, and noise level. Scene Labeling: Each image is annotated with the accident type (e.g., arc discharge, wire breakage) and severity (minor / moderate / major).

[0033] Step 2: Loss function design (multi-objective collaborative optimization).

[0034] The training loss of the gated network consists of three parts, which together guide weight learning: 1) Content-aware loss (Loss_content): uses the feature space constraints of the pre-trained VGG network to force the output image to be aligned with the expert annotations at the semantic level. .

[0035] Where Φ is the VGG feature extractor, which ensures that the enhanced image is consistent with the annotation in terms of texture, edges and other details.

[0036] 2) Branch-specific loss (L_branch): Apply L1 constraints to the output of each branch to prevent the degradation of professional ability: .

[0037] For example, in an electric arc scenario, the output of the brightness branch should be close to the expert-labeled brightness map, while the contrast branch needs to preserve the wire edge details.

[0038] 3) Gating regularization term (L_gate): Introduce scene prior constraints on weight distribution. For example, in arc discharge scenarios, the brightness branch weight should dominate: .

[0039] in is the scene-related weight prior (such as arc scene ), λ is the regularization strength coefficient.

[0040] Step 3: Curriculum-based training strategy (gradual capacity improvement).

[0041] 1) Stage 1 (pre-training): The gating network weights are fixed, and each branch network is optimized independently. The brightness branch trains an asymmetric mapping function to separate the arc glare from the background. The contrast branch adaptively adjusts the CLAHE parameters to enhance the edges of the wires. The electromagnetic denoising branch learns a residual mapping to suppress electromagnetic noise.

[0042] 2) Phase 2 (Joint Training): Enabling the gating network and performing end-to-end fine-tuning with a small learning rate (1e-4), the total loss is: .

[0043] 3) Stage 3 (Adversarial Training): A discriminator network is introduced to identify the authenticity of enhanced images, improve the naturalness of the output, and prevent overfitting.

[0044] (3) Dynamically generate fusion weights: Based on the learned fusion strategy, the gating network dynamically generates a fusion weight for each branch. These weights reflect the contribution of each branch in enhancing the image.

[0045] The process of dynamically generating fusion weights is as follows: 1) Feature extraction: The gating network first receives the image features processed by the three branches (brightness, contrast, and electromagnetic denoising). These features contain the different enhancement effects of each branch on the image.

[0046] 2) Understanding the scene: By analyzing these features, the gating network determines which accident scene the current image belongs to (such as arc discharge or wire breakage).

[0047] 3) Calculating Weights: Based on the identified scenario, the gating network assigns an importance score (weight) to each branch. For example, in an arc discharge scenario, the brightness branch responsible for processing strong light will receive a higher weight.

[0048] 4) Adjusting Weights: To make the weight distribution more reasonable, the network uses a "fine-tuning coefficient" to adjust these scores. Initially, the coefficient is large, allowing the network to explore more possibilities. As training progresses, the coefficient decreases, making the weight distribution more stable.

[0049] (4) Weight normalization: To ensure the quality of the fused image, the gating network uses the Softmax function to normalize the generated fusion weights. This way, the sum of the weights of all branches is 1, ensuring the stability of the fusion process.

[0050] (5) Weighted fusion of each branch output: Finally, the gating network performs a weighted summation of the outputs of each branch based on the normalized fusion weights. This results in a final enhanced image that incorporates the advantages of each branch while avoiding parameter conflicts and redundancy.

[0051] 4. Lightweight real-time processing module: During feature extraction, depthwise separable convolution is used to replace standard convolution to meet the real-time reasoning requirements of the drone end.

[0052] The key points and protection points of the present invention are: 1. Electromagnetic noise modeling and suppression: For the first time, frequency domain filtering and adaptive convolution are integrated in the preprocessing stage to specifically address stripe noise in high-voltage scenarios.

[0053] 2. Scene-adaptive dynamic attention mechanism: Automatically assign weights to each branch based on arc intensity and smoke concentration.

[0054] 3. Multi-task gated fusion: The brightness, contrast, and denoising branches are dynamically fused through a gating network to avoid parameter conflicts and improve robustness in complex scenes.

[0055] 4. Power feature enhancement module: Gradient-guided convolutional layers designed specifically for wires and insulators enhance device edge features.

[0056] The effects of the present invention are: 1. Noise suppression: The peak signal-to-noise ratio (PSNR) of streak noise is improved by ≥8dB (traditional filtering methods only improve it by 3-4dB).

[0057] 2. Fault detection accuracy: The arc recognition rate is increased to 98.7%, and the wire break positioning error is less than 0.5 pixels.

[0058] 3. Real-time performance: After lightweight design, the 1080P image processing delay is less than 50ms (U-Net enhanced model delay is greater than 120ms).

[0059] 4. Scenario adaptability: In the arc-smoke coexistence scenario, the structural similarity (SSIM) index is improved by 23% compared with the comparative patent model.

[0060] An embodiment of the present invention provides an image quality enhancement system, comprising: The image acquisition module is used to obtain the feature map of the high-voltage power line accident scene picture.

[0061] The branch extraction module is used to separate the arc glare and background in the feature map through an asymmetric mapping function to obtain the first feature map; enhance the image contrast of the feature map through the CLAHE algorithm and gradient guidance to obtain the second feature map; and learn the noise distribution in the feature map through residual learning and use frequency domain masking to denoise and suppress smoke artifacts to obtain the third feature map.

[0062] The image enhancement module is used to assign weights to the first, second and third feature maps and fuse them to obtain a quality-enhanced image.

[0063] A specific embodiment is as follows: This embodiment discloses a method for enhancing image quality, and the specific steps are as follows: S1. Arc discharge scene enhancement: Input: Arc image taken by a drone (strong light overexposure + electromagnetic stripe noise).

[0064] Processing flow: The electromagnetic front end suppresses noise (frequency domain filtering + adaptive convolution); dynamic attention focuses on the discharge point (channel weight increased to 0.9); the brightness branch separates the arc light, and the contrast branch enhances the wire texture.

[0065] Effect: The arc outline is clear, the background equipment details are discernible, and the fault point location time is shortened by 60%.

[0066] S2. Nighttime fog environment enhancement: Input: Image of a broken wire obscured by smoke at night.

[0067] Processing flow: Denoising branch suppresses smoke (residual learning + frequency domain masking); dynamic attention strengthens broken edges (spatial weight increased to 0.85); gated network fuses multi-branch results (brightness weight 0.3, contrast weight 0.6).

[0068] Effect: The accuracy of breaking point positioning reaches 99%, and the misjudgment rate is less than 1%.

[0069] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A method for enhancing image quality, characterized in that: The following steps are involved: Obtain feature maps of high-voltage power line accident scene images; The first feature map is obtained by separating the arc glare and background in the feature map through an asymmetric mapping function. The second feature map is obtained by enhancing the image contrast of the feature map through the CLAHE algorithm and gradient guidance. The noise distribution in the feature map is learned through residual learning and denoised and smoke artifacts are suppressed using frequency domain masks to obtain the third feature map. The first, second and third feature maps are weighted and fused to obtain a quality enhanced image.

2. The image quality enhancement method according to claim 1, wherein: Before obtaining the feature map of the high-voltage power line accident scene picture, the method further includes: pre-processing the captured high-voltage power line accident scene picture; The pre-processing step includes: using Fourier transform to convert the high-voltage power line accident scene picture from the time domain to the frequency domain, and filtering the frequency domain to remove noise; The filtering result is restored from the frequency domain to the time domain using inverse Fourier transform to obtain a filtered time domain image; The filtered time domain is input into the adaptive convolution, and the convolution kernel parameters are dynamically adjusted according to the noise intensity to suppress the residual noise.

3. The image quality enhancement method according to claim 2, wherein: The quality-enhanced image is obtained using an image enhancement network, which includes: a dynamic multi-task encoder, configured to input a pre-processed high-voltage power line accident scene image into the dynamic multi-task encoder to obtain a feature map of the high-voltage power line accident scene image; The brightness branch, contrast-texture branch, and electromagnetic-denoising branch are used to extract features from the feature maps of the high-voltage power line accident scene image to obtain the first, second, and third feature maps respectively; The gating network is used to input the first, second and third feature maps into the trained gating network for fusion to obtain a quality enhanced image.

4. The image quality enhancement method according to claim 3, wherein: The step of inputting the pre-processed high-voltage power line accident scene picture into a dynamic multi-task encoder to obtain a feature map of the high-voltage power line accident scene picture specifically includes: Based on the pre-processed high-voltage power line accident scene images, a lightweight convolutional neural network is used to extract the illumination and noise distribution features. Dynamically adjust the channel attention and spatial attention weights to enhance the fault area; according to the channel attention, adjust the feature strength of each channel through multiplication operation; according to the spatial attention, highlight the fault area through element-level multiplication or weighted summation; Obtain feature maps of high-voltage power line accident scene images.

5. The image quality enhancement method according to claim 3, wherein: The trained gating network specifically includes the following training steps: Obtain images of high-voltage power line accident scenes under different accident types; The gated network is trained using images of high-voltage power line accident scenes under different accident types to automatically adjust the weights of the brightness branch, contrast-texture branch, and electromagnetic-denoising branch in the fusion process. According to the fusion weight of each branch, the feature map of each branch is weighted summed to obtain the optimal fusion strategy under different accident types.

6. An image quality enhancement system, characterized in that: include: An image acquisition module is used to obtain a feature map of a high-voltage power line accident scene; A branch extraction module is used to separate the arc glare and background in the feature map through an asymmetric mapping function to obtain a first feature map; and to enhance the image contrast of the feature map through the CLAHE algorithm and gradient guidance to obtain a second feature map; The noise distribution in the residual learning feature map is used to denoise and suppress smoke artifacts using frequency domain masking to obtain the third feature map; The image enhancement module is used to assign weights to the first, second and third feature maps and fuse them to obtain a quality-enhanced image.

Images