摘要
Abstract
Addressing the challenges of inadequate real-time performance,high miss rates,and elevated false alarm rates in existing methods for detecting and identifying inland river vessels in complex environments,an enhanced YOLOv7-tiny-based method for ship detection and identification was proposed.The anchor frame was established through the K-means++clustering algorithm,effectively addressing the issue of preset anchor frames being ill-suited for detecting the extreme size variations of ships.Additionally,the SiLU activation function wass incorporated to replace the original model's LeakyReLU,optimizing the network architecture and bolstering the model's feature extraction capabilities.Introducing the GSConv lightweight convolution module into the ELAN module enriched feature information,facilitating the algorithm's deployment on lightweight mobile devices.The Efficient Multi-scale Attention(EMA)mechanism was integrated into the backbone feature extraction network,thereby enhancing the model's ability to perceive target location information across large multi-scale disparities.This model achieved a recognition accuracy of 97.2%on the SeaShips dataset,which is a 2.9%improvement over the previous version,with a 16.3%reduction in model size.Compared to Faster R-CNN,YOLOv5,and YOLOv7,the proposed method shows improvements of 7.1%,3.6%,and 1.0%,respectively.Experimental results demonstrate that the proposed algorithm excels in both accuracy and robustness.关键词
船舶检测/YOLOv7-tiny/K-means++聚类/SiLU/GSConv卷积/EMAKey words
ship detection/YOLOv7-tiny/K-means++clustering/SiLU/GSConv convolution/EMA分类
信息技术与安全科学
