Deep Interpretable Component Decoupled Dictionary Neural Network for Image Denoising in Industrial Cyber-Physical System

Image denoising techniques are pivotal in preprocessing noisy images, greatly enhancing the quality of visual data in applications within the realm of Cyber-Physical Systems (CPS). Take scenarios like autonomous vehicles and surveillance systems, for instance, where denoising plays a pivotal role in significantly improving the accuracy of object detection and recognition. However, the adoption of image denoising tasks in CPS is hindered by the fragility, robustness, and interpretability issues associated with neural networks. To address these challenges, this study introduces an innovative and interpretable approach to image denoising. We propose an image denoising model that combines dictionary learning with a deep neural network. This hybrid approach leverages decoupling and sparse convolution techniques, strategically designed to mitigate model fragility and reinforce model robustness. Furthermore, our model is geared towards untangling and reducing redundancy across different image components. The architecture of the network is crafted as a model-data-driven neural network, facilitating the simultaneous learning of various image components and deploying fusion mechanisms to mitigate perturbations and noise. Finally, we provide a theoretical framework to explain our method and substantiate its effectiveness through rigorous experimentation and validation.