UWaterDetect: A Lightweight Underwater Object Detection Model based on RT-DETR and Wavelet Convolutions

The advent of RT-DETR marks a pivotal advancement in object detection research. Despite their advancements, DETR-based detection frameworks remain underutilized in underwater scenarios due to their computational intensity and massive parameter requirements. To address this challenge, we design a parameter-efficient RT-DETR adaptation for underwater detection, achieving significant reductions without accuracy loss. Furthermore, we introduce a novel backbone, ResWT, which reduces parameters by 25.5% compared to the original network while maintaining model performance. Additionally, we designed the HLCG Hybrid Encoder, specifically optimized for small object detection, which significantly enhances accuracy in recognizing small objects. At the same time, our approach achieves a 19.4% reduction in parameters and an 18.1% drop in GFLOPs relative to the baseline RT-DETR.