Subsea Nodule Recognition and Deployment Detection Method Based on Improved YOLOv8s

An improved small-target detection model based on YOLOv8s is proposed to address the challenges associated with deep-sea polymetallic nodule detection, such as complex target shapes, small sizes, and strong environmental interference. This model is capable of real-time deployment on the Orange Pi 5 development board and is designed to efficiently and accurately identify seafloor polymetallic nodules. The enhanced model incorporates an adaptive image preprocessing module and an EMA attention mechanism into the YOLOv8s framework. Additionally, SK-Conv convolution modules replace certain Conv and C2f layers in the backbone network, while the Focal SIoU loss function replaces the original CIoU loss function. These modifications enhance feature extraction capabilities in the presence of uneven lighting and background interference, optimizing nodule segmentation in complex backgrounds and improving small-target detection performance. Experimental results demonstrate that the improved YOLOv8s model offers significant advantages in terms of detection accuracy and robustness, achieving precision, recall, and mean average precision rates of 94.88%, 93.89%, and 96.2%, respectively. The deployed YOLOv8s-OR model strikes a balance between recognition accuracy and speed by maintaining an average frame rate of 37.5 FPS with a latency of 22 ms, providing an efficient and precise solution for automated deep-sea polymetallic nodule detection with practical value.