Water Surface Target Detection Algorithm for Unmanned Cleaning Ship Based on Improved YOLO V5

Abstract

The appearance of the unmanned cleaning ship has improved the efficiency and safety of water surface cleaning. Accurate water surface target recognition is the premise for the unmanned cleaning ship to complete the cleaning inspection task. However, the current water surface target detection algorithm is not accurate enough in small target detection because of its network structure is lack of optimization in small target detection, which affects the completion of the water surface cleaning task. In this study, the improved YOLO V5 algorithm is used for water surface target detection. The C3 structure in YOLO network is replaced by the encoder structure in transformer network, and the small target detection layer and the Convolutional Block Attention Module (CBAM) are integrated. Finally, this study constructs water surface target detection datasets, and test the processed datasets using the improved YOLO V5 algorithm. The experimental results show that the improved surface target detection algorithm has better performance and higher average detection accuracy in small target detection, which proves that the method has certain practicability and progressiveness in the application of clean unmanned ship.