基于Hadoop的煤矿图像PCA-SIFT特征提取算法研究  

PCA-SIFT Feature Extraction Algorithm of Coal Mine Image in Hadoop Platform

作　　者：米向荣[1] 曹建芳[1] 史昊[2] MI Xiang-rong;CAO Jian-fang;SHI Hao(Computer Department, Xinzhou Teachers’ University, Xinzhou 034000, China;College of Computer Science and Technology, Taiyuan University of Science and Technology, Taiyuan 030024, China)

机构地区：[1]忻州师范学院计算机系,山西忻州034000 [2]太原科技大学计算机科学与技术学院,山西太原030024

出　　处：《软件导刊》2018年第12期81-86,共6页Software Guide

基　　金：山西省自然科学基金项目(2014011019-3);山西省科技重大专项项目(20121101001);山西省-中科院科技合作项目(20141101001)

摘　　要：大数据技术已经成为当下热点问题,Hadoop技术在煤矿领域运用也引起了广泛关注。针对传统监控模式下煤矿视频监控系统图像采集点多、历史留存数据量大、不利于后续查找特征图像等问题,提出一种Hadoop平台下PCA-SIFT算子的图像特征提取算法,研究并改进了MapReduce并行编程模型的任务设计,对传统尺度不变特征转换算法进行了并行化设计,在Hadoop集群下实现了海量煤矿图像的PCA-SIFT并行特征提取。使用汾西矿务局煤矿图像井下数据集进行实验,算法SIFT特征点检测效果好,运行耗时少。在图像数量庞大时,系统加速比几乎呈线性增长趋势,验证了算法处理大规模煤矿图像数据的有效性。Large data technology has become a hot issue at present,and Hadoop technology has also attracted widespread attention in the field of coal mining.Aiming at multiple the image acquisition points of traditional video monitoring system of coal mine monitoring mode and the huge historical data which are not conducive to the subsequent search features of images and other issues,this paper presents an image feature of a Hadoop platform for the optimization of SIFT operator algorithm.We studied and improved design of MapReduce parallel programming model and the traditional scale invariant feature conversion algorithm for parallel design in the Hadoop cluster and implemented parallel SIFT image feature extraction of massive coal mine.By using the Pascal VOC2012data set for experiment,the effect of the SIFT feature point detection algorithm is proposed and the operation is less time-consuming.When dealing with a large number of images,the system speedup is almost linearly,which verifies the validity of data processing algorithm of large-scale coal mine image.

关 键 词：SIFT算子 尺度不变特征 HADOOP平台 MapReduce并行编程模型 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]