Hybrid U-ResNet-Dense Image Denoising for Low-Light Stellar Cameras
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image denoising algorithms fail to achieve effective denoising performance while preserving image quality, particularly in low-light conditions, affecting the visual effect and accuracy of object recognition in images captured by stellar cameras.
Innovation Solution
An image denoising method utilizing a model combining a U-shaped network, a residual network, and a dense network to enhance the removal of noise while preserving image details, trained using an adversarial network to generate additional training samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If existing image denoising algorithms are used, then noise removal is achieved, but image quality and details are lost
Solution Approach 1:
The patent combines three different network architectures (U-shaped network for multi-scale feature extraction, residual network for gradient flow and detail preservation, and dense network for feature reuse) into a unified hybrid model. This merging allows the system to simultaneously achieve effective noise removal while preserving image quality and details, resolving the contradiction between noise reduction and quality maintenance.
Solution Approach 2:
The hybrid network model functions as a composite structure, integrating different neural network components (U-shaped, residual, and dense blocks) with complementary strengths. Each component contributes specific capabilities: U-shaped blocks for contextual understanding, residual blocks for gradient flow, and dense blocks for feature propagation, creating a robust composite system that maintains image quality while removing noise.
2Illumination intensity
If stellar cameras are used in low-light conditions, then monitoring capability is improved, but image resolution and recognition accuracy deteriorate
Solution Approach 1:
The patent converts the harmful effect of low-light noise into a training opportunity by using adversarial networks. The generator network learns to produce realistic clean images from noisy low-light inputs, while the discriminator network identifies authentic image characteristics. This adversarial training transforms the low-light noise problem into a benefit, enabling the model to enhance recognition accuracy in previously problematic conditions.
Solution Approach 2:
The model dynamically adjusts processing parameters based on input image characteristics, particularly adapting to low-light conditions by modifying feature extraction weights and noise filtering强度. The adversarial training process learns optimal parameter configurations for different lighting conditions, enabling the system to maintain high recognition accuracy across varying illumination levels.
3Productivity
If conventional denoising methods are used, then processing speed is maintained, but denoising performance and generalization capability are insufficient
Solution Approach 1:
The patent implements comprehensive pre-training using adversarial networks to generate diverse training samples before actual denoising operations. The model undergoes extensive training on synthetic and real noisy images, learning robust noise patterns and cleaning strategies in advance. This preliminary action ensures that when the model processes actual images, it can quickly apply learned denoising patterns without compromising speed, while achieving superior performance and generalization.
Data Source
AI summary
The present application relates to the field of image processing, and provides an image denoising method and apparatus, an electronic device and a storage medium. The image denoising method includes: acquiring an image to be processed, and inputting the image to be processed into an image denoising model to acquire a denoised image, wherein the image denoising model is a model formed by combining a U-shaped network, a residual network and a dense network.