YOLOv7-AD: An Efficient and Accurate Underwater Object Detection Algorithm Based on YOLOv7

Underwater target detection presents challenges such as intricate backgrounds, image blurriness, and the process of feature extraction is rendered more complex and susceptible to external interference when the target clustering is small. In light of this, it is essential to refine the algorithm optimization approach to ensure its integrity and accuracy of the extracted information, effectively reduce adverse factors such as background noise, and enhance the precision of detection and the overall resilience of the system. Advanced algorithms have been incorporated to enhance accuracy, accelerate processing speed, and reduce resource consumption through a lightweight design, thereby ensuring efficient operation and straightforward maintenance. Built upon YOLOv7, YOLOv7-AD employs a disconnected head configuration that autonomously handles classification and regression functions, thereby improving feature extraction. It adopts an anchor-free design to simplify the detection process, reduce prediction time, and improve adaptability in underwater environments. Furthermore, the integration of the CBAM attention mechanism boosts feature representation across both channel and spatial dimensions. Comprehensive evaluations reveal that YOLOv7-AD outperforms YOLOv7 with a 1.23% increase in average detection accuracy (mAP), a 1.52% boost in inference speed (FPS), and a 6.78% reduction in model parameters.