An Improved YOLO v3 Small-Scale Ship Target Detection Algorithm
Haiyan Yu, Yu Li, Dexian Zhang
Abstract
Because of the problems of missed detection, false detection and low accuracy of Yolo V3 algorithm in small-scale ship image, an improved Yolo V3 ship target detection algorithm is proposed and applied to small-scale ship image detection. Firstly, the small-scale feature layer is fused with the second and first feature layers, and the enhanced feature layer is output. Then, according to the characteristics of the ship, the length width ratio is added to the loss function to make the loss function more suitable for the ship image. Finally, the improved Yolo V3 algorithm is used to train the data model. Experimental results show that the improved Yolo V3 algorithm can effectively detect small-scale ship targets, and the recall rate and accuracy rate are improved compared with the original algorithm, which can meet the needs of fast and accurate ship detection.