Underwater Object Detection Algorithm Based on Improved YOLOv5

Underwater object detection is a crucial aspect of autonomous underwater vehicle (AUV) sensing systems, playing a vital role in ocean exploration and environmental monitoring. However, the unique characteristics of underwater environments bring forth various challenges. These challenges include object ambiguity, limited visibility, small object size, and similarity to the background, making target labeling and dataset curation highly complex. Consequently, datasets in this domain tend to be both small in size and unbalanced in terms of class distribution. To address these challenges, this paper presents an enhanced approach that incorporates a cross attention (CA) module for global feature extraction and establishing correlations. Additionally, a texture enhancement (TE) module is integrated into the neck of the network to extract target edge information, facilitating better separation of underwater targets from the background. Furthermore, an Adaptive Weighted Fusion (AWF) module is introduced in the detection layers to improve decision consistency. Through extensive experimentation on the URPC2022 dataset, the effectiveness of our proposed method is demonstrated, showcasing its ability to tackle the aforementioned challenges in underwater object detection.