YOLO26-DeepSpace:一种面向深空弱小目标的多尺度多目标检测跟踪网络OA

For space early warning and space target surveillance systems,aiming to effectively detect various space debris and other small-size,weak-signal celestial targets,this paper proposes a target detection and tracking algorithm based on improved YOLO26 to address the challenges in deep space environments such as noise interference,faint target features and discontinuous trajectories.Firstly,an image sub-region self-perception enhancement module is deployed to enhance the visual clarity of dim and dark targets via local contrast optimization and edge preservation mechanism.Secondly,the YOLO26 detection network is lightweighted,and an improved multi-scale attention module GCEMA is introduced to strengthen the detection capability for dim and small targets.Finally,combined with the OCSort tracking algorithm and the appearance re-identification model CosmicReID,multi-target cataloging and stable tracking are realized.Experimental results show that compared with the original YOLO26 network and OCSort algorithm,the improved network has fewer parameters and faster inference speed,with the detection speed increased by 11.9%;the tracking accuracy of the integrated system is improved by 6.2%,and the frequency of target identity switching is reduced by 48.5%.The proposed method meets the requirements on detection accuracy and tracking accuracy of automatic tracking and measurement systems.