关键词: 船舶识别;目标检测;深度学习;卷积神经网络;YOLOv4-Tiny;

Abstract: In order to effectively improve the accurate and rapid detection of ships on rivers and lakes in complex river-lake environment, an object detection algorithm for ships on rivers and lakes based on improved YOLOv4-Tiny was proposed.The algorithm first activated the function by introducing a Sigmoid Weighted Linear Unit(SiLU),then the Convolution+Batch Normalization+SiLU(CBS)was constructed and replaced the original module(Conv+BN+Leak＿relu)in the backbone network.Then, an improved Spatial Pyramid Pooling(SPP)network was added after the backbone network.Finally, a Convolution Block Attention Module(CBAM)was introduced into the Feature Pyramid Network(FPN),and a top-down connection was adopted, an improved Path Aggregation Network(PANet)was built.The experimental results showed that the average accuracy of the improved YOLOv4-Tiny target detection algorithm was 1.27% higher than that of the original algorithm in the complex environment of darkness, blur, strong light and shelter overlap.The detection speed reached to 81.55 FPS,and the occupied memory was only 37.34 MB.The research results can provide a reference for the intelligent management of ships passing through rivers and lakes.

Key words: ships detection; object detection; deep learning; convolutional neural network; YOLOv4-Tiny;