污染修复场地中基于显著性的目标检测方法  

Detection method of object based on saliency in polluted remediation site

作　　者：张宝峰[1,2] 曹珍珍 朱均超 刘娜[1,2] ZHANG Baofeng;CAO Zhenzhen;ZHU Junchao;LIU Na(Tianjin Key Laboratory of Complex System Control Theory and Application,Tianjin University of Technology,Tianjin 300384,China;School of Electrical and Electronic Engineering,Tianjin University of Technology,Tianjin 300384,China)

机构地区：[1]天津理工大学天津市复杂系统控制理论及应用重点实验室,天津300384 [2]津理工大学电气电子工程学院,天津300384

出　　处：《天津理工大学学报》2021年第3期55-59,共5页Journal of Tianjin University of Technology

基　　金：天津市互联网跨界融合创新科技重大专项(18ZXRHSF00240)。

摘　　要：为解决污染修复环境中施工人员实时防护问题,提出使用基于改进的流形排序显著性检测算法对施工人员进行检测。首先,利用超像素分割对图像进行处理,建立适合的图模型;然后利用流形排序方法将图像各边界节点作为背景种子得到第一个显著图;再利用流形排序对第一阶段的显著图进行阈值分割,得到最终的显著图。实验表明,改进方法在视觉效果上与传统检测方法相比,得到的显著区域有很好的改善,并且在精确率、召回率、F值等评价指标上有较好的反映。In order to solve the problem of real-time protection of construction personnel in the environment of pollution remediation.This paper proposes to use the improved salience detection algorithm based manifold ranking to detect the construction personnel.Firstly,super-pixel segmentation is used to process the image and establish a suitable image model;Then the nodes on the image boundary are used as background seeds by manifold ranking algorithm;Finally,the first stage saliency map is segmented by using manifold sorting to get the final saliency map.Experimental results show that the improved method is better than the traditional detection method in visual effect,and has a better reflection on the precision rate,recall rate,F-measure and other evaluation indicators.

关 键 词：污染修复环境 流形排序 显著性检测 超像素分割 

分 类 号：TV139.2[水利工程—水力学及河流动力学]