摘要
为了降低伪目标引起的误检率,提高系统在复杂环境中的目标识别能力,设计了一种基于立体视觉分析的光谱聚类算法,该算法在结合待测目标几何特性的基础上完成光谱聚类分析,从而实现通过立体视觉作为边界条件的方式消除伪目标的干扰。实验采用TEL-2000型成像光谱仪采集的目标区域图像作为样本与检测数据,分别对不同条件下的目标光谱特征值、非目标光谱特征值以及相关系数进行检测分析,对比目标与伪目标的识别效果。结果显示,目标光谱特征参数个数越多,目标检出概率越大,但伪目标误检概率也较大;非目标光谱特征参数越多,伪目标误检概率越小,但目标检出概率降低;当u=6,v=4,η=0. 6时,识别效果最好。该算法能够保证高目标检出率时实现误检率的有效降低值。
In order to reduce the probability of false detection caused by false targets and improve the target recognition ability of the system in complex environment,a spectral clustering algorithm based on stereo vision analysis is designed.Based on the geometric characteristics of the target to be measured,spectral clustering analysis is completed.Stereo vision acts as a boundary condition to eliminate the interference of the false target.The TEL-2000 imaging spectrometer is used in experiments to collect image information of the target area as sample and test data.The spectral characteristic values of target are detected and analyzed under different conditions.The spectral characteristic values of target and background are given,and the recognition effects are compared.The results show that the more the target spectral feature parameters,the greater the target detection probability,but the probability of false target is also larger;The more non-target spectral feature parameters,the smaller the probability of false target,but the probability of target decreases;When u=6,v=4,h=0.6,the recognition effect is the best.The algorithm can effectively improve the probability of detection,and reduce the probability of false target.
作者
韩伟佳
王国伟
孙亚东
李超然
HAN Weijia;WANG Guowei;SUN Yadong;LI Chaoran(College of Information Technology,Jilin Agricultural University,Changchun 130118,China;Information Engineering College,College Of Optical and Electronical Information,Changchun University of Science and Technology,Changchun 130114,China)
出处
《激光杂志》
北大核心
2019年第5期23-26,共4页
Laser Journal
基金
吉林省自然科学基金
关键词
目标识别
谱聚算法
立体视觉分析
识别概率
target recognition
spectral clustering algorithm
stereo vision analysis
recognition probability
