Research On Maritime Ship Identification Based On Deep Learning

With the rapid development of maritime trade,the importance of maritime safety is increasingly being valued.Realizing intelligent recognition of maritime ship targets not only aims to safeguard China’s marine resources,but also plays an important role in efficient marine production,military activities,and other aspects,which promotes the unmanned and intelligent development of ships.Existing target recognition algorithms are often limited by the weather and background they are in when performing image recognition.If the weather is harsh or the background is complex,it will affect the detection effect and quality of the target.Against the backdrop of the flourishing development of deep learning,object detection methods based on deep learning have been widely applied due to their outstanding characteristics.This article proposes a maritime ship recognition algorithm based on improved YOLOv5 deep learning to address the complexity of the marine environment and the low recognition rate of ship recognition algorithms.The main research content of this article is as follows:(1)Firstly,the development history of deep learning at home and abroad,the current status of ship recognition research,and related theories are introduced.Then,three representative object detection algorithms are listed and analyzed.Finally,the YOLOv5 algorithm used in this article is proposed.(2)Research on maritime ship recognition based on deep learning.During image input,the effective information of the ship image is enhanced by processing the image.In order to make the model pay more attention to the feature information of learning channels and integrate more features,the SE attention mechanism module is added to YOLOv5.Aiming at the problem of ship occlusion during the detection process,NMS was improved to improve the recall ability of the target ship.(3)Conduct experimental verification on different algorithms.In ship identification tasks,the accuracy,recall,and F1 values of the improved YOLOv5 model were 90.6%,89.9%,and90.5%,respectively.The detection effect was improved by 6.3%,4.8%,and 5.8% compared to YOLOv5 model,and 19.1%,19.0%,and 19.3% compared to YOLOv4 model,respectively;In the task of ship identification in foggy days,the accuracy,recall,and F1 values of the dark channel defogging algorithm model were 88.1%,87.2%,and 87.6%,respectively.The detection results were 13.8%,13.3%,and 13.5% higher than those of the non-fogging algorithm model,respectively.The research results show that the improved YOLOv5 model effectively solves the problem of low accuracy in marine ship detection under multi target and foggy conditions and improves the overall effect of ship recognition.