基于数据挖掘的偏振光成像目标检测研究  

Research on polarized light imaging target detection based on Data Mining

作　　者：张樊[1] 王晶[1] ZHANG Fan;WANG Jing(School of information engineering,Wuchang Institute of Technology,Wuhan 430065,China)

机构地区：[1]武昌工学院信息工程学院,武汉430065

出　　处：《激光杂志》2023年第12期225-230,共6页Laser Journal

基　　金：湖北省教育厅科学研究计划指导性项目(No.B2021330)。

摘　　要：为提升偏正光图像目标识别技术水平,研究基于数据挖掘的偏振光成像目标检测方法。使用阈值分割法对偏振光图像实施分割预处理;使用斯托克斯参量、方差等方法获取偏振光图像的偏振态特征、目标边缘特征和直线特征;利用深度学习神经网络模型输出偏振光成像目标检测结果。实验结果表明:该方法可有效分割偏振光图像内的目标和背景,提取偏振信息能力较好;提取偏振光图像特征像素点的最低精度为0.98左右,提取其特征能力较强;特征信息提取效率较高,对于偏振光图像的特征信息提取最低耗时为0.257 s,应用性能较好。In order to improve the technology level of polarizing image target recognition,a polarizing image target detection method based on data mining is studied.The threshold segmentation method is used to implement the seg-mentation preprocessing of the polarized light images;the polarization features,target edge features and line features of the polarized light images are obtained by using Stokes parameters and variance methods;the deep learning neural net-work model is used to output the detection results of polarized light imaging targets.The experimental results show that this method can effectively segment the target and the background within a polarized light image,and has a good ability of extracting polarization information;the minimum accuracy of extracting feature pixels of polarized light image is a-bout 0.98,and its feature extraction ability is strong;the feature information extraction efficiency is high,and the minimum time of feature information extraction for polarized light image is 0.257 s,and the application performance is good.

关 键 词：数据挖掘 偏振光成像 目标检测 图像分割 偏振信息模型 深度学习 

分 类 号：TN209[电子电信—物理电子学]