Water surface multi-target detection method of intelligent ship water surface based on improved YOLOv8

Objectives In order to improve the stability and accuracy of intelligent ships in the automatic detection of floating objects on the water surface Methods a water surface multi-target detection method of intelligent ship based on improved YOLOv8 algorithm is proposed. In order to enhance the detection ability of the model for objects of different scales, DDCNv4 module is designed to replace C2f module in Backbone network. At the connection with the Neck layer, the Bi-Level Routing Attention mechanism is introduced to strengthen the performance of the model in anti-interference and texture feature extraction, and optimize the detection effect of small-size objects. In addition, the DSConv module is further integrated in the Neck layer to realize the lightweight of the model, which takes into account the model performance and computational efficiency. Results The experimental results show that compared with the traditional YOLOv5, YOLOv7, YOLOv8 and other methods, the detection accuracy of the improved YOLOv8 algorithm proposed in this study reaches 90.4%, which is 5.6% higher than the mAP@0.5 of the original YOLOv8 algorithm, and the number of parameters is reduced by 0.1M. The model inference speed is improved by 1.98FPS. Conclusions The algorithm can maintain a high level of detection performance in the complex water surface environment, which provides theoretical reference for the application of target detection in the field of intelligent ship.