Research and application of Yangtze River fishing boat identification during fishing ban based on improved YOLOv5s algorithm
The Ten-year Fishing Ban of the Yangtze River is a critical component of the river's ecological restoration efforts.To address challenges such as small target sizes,complex backgrounds,and high mobility of illegal fishing vessels during the ban,an improved target detection algorithm based on YOLOv5s has been developed.This enhanced algorithm optimizes a multi-scale adaptive anchor module and substitutes the original K-means clustering algorithm with an advanced K-means++clustering algorithm,redefining the clustering to match the specific dimensions of the Yangtze River vessels.At the output end of the backbone network,a lightweight and efficient coordinate attention(CA)mechanism is introduced,reducing background noise interference and focusing on identifying key features of vessels,which diminishes computational and parameter requirements,thereby boosting detection efficiency.The algorithm utilizes an enhanced spatial pyramid pooling and context-aware spatial pyramid pooling combination(SPPCSPPC)module for pooling feature maps,which enriches the target detection network's ability to perceive and express multi-scale features,notably improving the detection capability for small-scale targets.Yangtze River vessel image data is collected and enhanced to construct a dedicated dataset for training to derive an optimal weight model.The final improved model has exhibited increases in accuracy,recall,mAPO.5,and mAPO.5∶0.95 by 1.5％,3.0％,2.4％,and 7.7％.The improved model also demonstrates a faster loss convergence during training,resulting in a lower final loss value.It surpasses other lightweight models and proves that the enhancements have effectively increased target detection accuracy while it also ensures efficiency.This makes it suitable for real-time detection requirements for the Yangtze River fishing vessels,providing technological support for the Ten-year Fishing Ban of the Yangtze River.
万方数据
维普
