A high signal-to-noise ratio descattering method on rainy via small data

Deep learning has achieved significant strides within the realm of image descattering tasks. However, the preponderance of extant methods are chiefly centered around the elimination of raindrops or rain streaks presenting a singular geometric configuration or characteristic form. Their capacity for generalization is markedly constrained when confronted with intricate real-world settings typified by the concurrent presence of rain. This paper presents a high signal-to-noise ratio descattering method on rainy via small data. This methodology adopts an all-in-one training paradigm entailing the incremental integration with small data of diverse rainy. Training datasets comprehensively encapsulate raindrops, rain streaks, and rain manifestations across a spectrum of distinct morphological features. Through this approach, the model is enabled to adapt in a more proficient and resilient manner to complex rainy environmental conditions. Experimental outcomes unambiguously evince that the method proposed herein attains exemplary descattering outcomes in authentic rainy scenarios. It achieves an improvement of 17.1% in the average signal-to-noise ratio and manifests enhanced generalization capability and augmented robustness.