Improved YOLO for Underwater Small Target Detection with New Detector Head and Attention Mechanism

Underwater small target detection in side scan sonar (SSS) imaging remains a difficult problem. In this paper, we propose an improvement method based on YOLO (You Only Look Once) by tackling two critical issues: preserving fine features of small targets and suppressing background interference. Our specific contributions are twofold: (i) we incorporate shallow layer feature maps to retain spatial details that are essential for small objects, and (ii) we introduce an attention mechanism into the feature fusion stage to improve discriminative ability against seafloor background. This improved method is verified by the public data collected by Portugal government, which achieves a precision of 83.6%, thereby enabling more accurate target detection. Experimental results proved the effectiveness of the proposed method, which can be applied in the underwater small target detection field.