采用聚合Gabor核和局部二元模式的烟雾识别方法  

Smoke Recognition by Combining Aggregated Gabor Kernels and Local Binary Patterns

作　　者：袁非牛[1,2] 李钢 夏雪 章琳[1,4] 周宇 YUAN Fei-niu;LI Gang;XIA Xue;ZHANG Lin;ZHOU Yu(School of Information Technology, Jiangxi University of Finance and Economics, Nanchang 330032, China;College of Information,Mechanical and Electrical Engineering,Shanghai Normal University,Shanghai 201418 ,China;College of Mathematics and Computational Science, Yichun University, Yichun 336000, China;School of Mathematics and Computer Science, Jiangxi Science and Technology Normal University, Nanchang 330038, China)

机构地区：[1]江西财经大学信息管理学院,南昌330032 [2]上海师范大学信息与机电工程学院,上海201418 [3]宜春学院数计学院,江西宜春336000 [4]江西科技师范大学数学与计算机科学学院,南昌330038

出　　处：《小型微型计算机系统》2019年第4期827-833,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61862029)资助;江西省高校科技落地计划项目(KJLD12066)资助;江西省教育厅科技项目(GJJ170317;GJJ150406;GJJ170892)资助;江西省科技支撑计划项目(2015ZBBE50013)资助;江西省研究生创新专项资金项目(YC2017-B070)资助

摘　　要：针对图像烟雾识别存在高误报率和高错误率的问题,提出一种新的结合聚合Gabor核和局部二值模式(LBP)的烟雾识别方法.首先使用加权平均法生成聚合Gabor核,以减少Gabor核的数目,并将聚合Gabor核和原图像进行卷积,生成聚合Gabor特征图.为了增强分类性能,本文提出了一个自定义比较函数,用于原图像和Gabor特征图的局部二值编码.然后生成3种LBP映射模式的特征,并串联这3种特征得到LBP串联模式特征.最后串联所有的LBP串联模式特征作为原图像的局部聚合Gabor二值模式(LAGBP)特征.与传统方法的比较实验显示,在3个测试集上,本文方法都分别要优于传统方法,且具有最低的错误率;与深度学习方法的比较实验中,本文方法也有很好的表现.实验表明LAGBP方法具有很好的烟雾特征表达能力,非常适用于烟雾识别.To reduce false alarm and error rates,we propose a novel smoke recognition method by combining aggregated Gabor kernels and Local Binary Patterns( LBPs). First,aggregated Gabor kernels are generated using the weighted sum of original Gabor kernels,and then aggregated Gabor feature maps are produced by convolving the aggregated Gabor kernels with an image. To enhance classification performance,we propose a self-defined comparison function,which is used to encode the original image and aggregated Gabor feature maps. Then we compute three histograms of each LBP code with three LBP modes,and concatenate them to produce an LBP mode connection histogram. Finally,we concatenate all LBP mode connection histograms to achieve a histogram of Local Aggregated Gabor Binary Pattern( LAGBP). Comparison experiments show that the proposed method outperforms traditional methods and has the lowest error rates among them on all test sets. In comparison experiments with several deep learning methods,our method also achieves an outstanding performance. Therefore,the proposed method has a good discriminative ability and is suitable for smoke recognition.

关 键 词：烟雾识别 Gabor核 LBP 聚合Gabor核 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]