基于YOLOv11-DeepSORT改进的遥感图像舰船多目标跟踪算法

打开文本图片集

中图分类号：TP391.41 文献标志码：A 文章编号：2095-2945（2025）28-0023-10

Abstract：Multi-object trackingofshipsinremotesensingimagesplaysacrucialroleinmarinemonitoring，environmental protectionandotherrelatedfelds.Currently，detection-basedmulti-objecttracking（MOT）methodsdominatethefield，where trackingperformanceheavilydependsontheacuracyofthedetectionnetworkToaddresstheisueoflowtrackingperfomance causedbyinsuffcientdetectionaccuracyinexistingalgorithms，thispaperproposesanimprovedYOLOv11objectdetection algorithmbasedontheRMS-FPNfeatureenhancementnetwork.Aditionally，LDConvisintegratedintothebackbonenetwork， andtwospecializedmodulesaredesignedforshipdetectioninremotesensing images：theSSFTandthe SDFT.Comparedtothe original YOLOv11， the improved method achieves an accuracy increase of 11.0% ，arecall improvement of 5.0% ，an mAP50 increase of 10.4% and an mAP50-95 increase of 5.2% .Furthermore，the feature extraction network ofDeepSORTisoptimized by replacing the original ResNet with EfcientNet，leading to a 29.92% improvement in MOTA，a 4.14% increase in IDF1，and a reduction of 53 IDscompared to the baseline.

Keywords： ship detection; ship tracking; YOLOv11; DeepSORT; feature enhanced networ

随着全球化的推进，海洋运输成为国际贸易的重要组成部分，舰船作为海洋活动的主要载体，对其进行有效检测和跟踪显得尤为重要。（剩余12829字）