Investigation of Complex Underwater Environments for Target Recognition Based on Improved YOLOv7-Tiny

The sustainable utilization of marine biological resources requires long-term monitoring of underwater ecosystems, yet traditional detection methods often rely on manual labor and fall short. This study addresses underwater target detection in complex environments, using scallop recognition as a case study, and proposes an improved YOLOv7-tiny algorithm. First, the Coordinate Attention (CA) mechanism is integrated into the neck component, resulting in a Coordinate Attention Convolution (CAConv) that enhances model performance. Second, the Feature Pyramid Network (FPN) structure is strengthened to improve detection across various scales. Finally, the EIoU loss function is incorporated to boost accuracy and robustness. Compared to similar detection models, the proposed algorithm shows significant improvements in average precision and recall rate.