Description of Invention
In this research, we propose, KBY-Net, a novel end-to-end Y-Net architecture built upon YOLOv8 that addresses the challenge of object detection in rainy weather conditions. It leverages multi-task learning for simultaneous image restoration and object detection. The network incorporates a KBY-decoder for image deraining, a transposed attention (MDTA) module for capturing long-range dependencies, and a multi-axis feature fusion (MFF) block for refined feature extraction. Evaluations on RainCityscapes and RAIN-KITTI datasets demonstrate that KBY-Net significantly outperforms state-of-the-art object detection approaches in challenging rainy scenes.