距离信息与排序模型融合的显著性目标检测方法  

A Salient Object Detection Method Fused with Distance Information and Manifold Ranking

作　　者：陈睿琦 南心蒙 常皓 CHEN Ruiqi;NAN Xinmeng;CHANG Hao(Xi’an Mordern Control Technology Research Institute,Xi’an 710065,Shaanxi,China;No.32382 Unit,Beijing 100072,China)

机构地区：[1]西安现代控制技术研究所,陕西西安710065 [2]32382部队,北京100072

出　　处：《弹箭与制导学报》2023年第3期33-38,共6页Journal of Projectiles,Rockets,Missiles and Guidance

摘　　要：针对图像探测系统对地面场景中目标探测及识别问题,提出基于距离轮廓信息与排序模型相融合的显著性目标检测方法,利用二维简易距离轮廓信息对可见光及红外图像实施分割,采用流形排序模型对图像进行处理,进而提高目标的检测性能。实验结果表明,白天条件下对显著性目标检测的准确率可达到87.7%,召回率达到91.6%;夜间条件下的检测准确率达到81.2%,召回率达到86.3%,证明了该方法在保证图像处理速度的前提下,能够显著提高目标检测的正确性。Aiming at the problem of detecting and recognizing targets from complex backgrounds,a salient object detecting method based on the combination of distance information and manifold ranking is proposed.In order to improve the detection performance in complex backgrounds,two-dimensional distance information is used to segment visible light images and infrared images,and manifold ranking is used to process the images.Experimental results show that under daylight condition the precision and recall of this method are 87.7% and 91.6% respectively;while under nighttime condition the precision and recall are above 81.2% and 86.3%.Results show the proposed method can significantly improve the accuracy of target detection from complex backgrounds while ensuring the speed of image processing.

关 键 词：图像处理 显著性目标检测 目标识别 流形排序 距离信息 

分 类 号：E91[军事] TJ765.3[兵器科学与技术—武器系统与运用工程] TP391.4[自动化与计算机技术—计算机应用技术]