Research On Ship Target Detection And Recognition Method Based On Improved YOLO Network

With the development of artificial intelligence technology,deep learning algorithms have been widely used in the field of object detection due to their powerful feature extraction ability to learn abstract and high-dimensional features from massive data.Deep learning algorithms have performed well in many fields,but their performance in specific ship detection tasks is not ideal.This article proposes an improved YOLO algorithm to address the above issues,with the specific content as follows:Firstly,in order to solve the problem of weak detection ability of YOLOv4-Tiny network,a new YOLOv4-A network was designed.Firstly,to address the problem of slow convergence speed in ship detection training,K-means++clustering algorithm is used to optimize the candidate box generation mechanism,perform clustering analysis on target labels,and accelerate training speed;In response to the problem that the original network’s FPN structure cannot fully integrate contextual information,a Bi FPN feature fusion mechanism is introduced to enhance the network’s ability to extract target feature information.In response to the issue of false detections caused by coastal interference during the detection process,an attention mechanism module has been introduced in the feature enhancement section to further enhance the network.Secondly,in order to enhance the detection performance of the network in various practical scenarios,this paper proposes an improved YOLOv4 network-M-YOLOv4,for real-time and efficient detection of ship targets.In response to the problem of slow feature extraction speed in Darket Net53 network,the Mobile Net V1 module is introduced to achieve lightweight backbone network;In response to the problem of high computational complexity in the 3 and 5 convolutional blocks in YOLOv4,which affects network detection,separable convolutions are introduced to greatly improve the detection speed of targets;In response to the fact that the non maximum suppression algorithm in the original YOLOv4 easily removes the correct prediction box when detecting occluded ship targets,resulting in missed detections,the adaptive non maximum algorithm is introduced to improve detection accuracy.The experimental results show that the algorithm proposed in this article performs well in ship detection,significantly improving detection accuracy and speed,especially in target detection of small and occluded ships.