基于随机森林分类的快速标签检测  

Fast Label Inspection Based on Random Forest Classifer

作　　者：章沛[1] 陈小瑜[1] 

机构地区：[1]广州医科大学信息技术中心,广州510182

出　　处：《重庆师范大学学报（自然科学版）》2015年第5期131-136,共6页Journal of Chongqing Normal University:Natural Science

摘　　要：主要解决了传统的人工标签检测中存在的成本高、速度慢以及检测率低的问题。与传统算法相比,提出了一种基于随机森林分类的快速标签检测算法。文章首先设计了自适应特征选择方法,能够在大规模的特征中自动选择鉴别力最高的特征,较好地区分前景和背景。在此引入了基于金字塔随机撒块的随机森林分类器,能够快速有效地完成多种标签的检测。实验结果验证了本算法的高速、高精度特性,适用于高速自动化生产线上的标签检测。Traditional manual inspection for label confronts problems such as high economic cost, low speed and low accuracy. Ai-ming at those, this paper proposes a quick inspection methods based on random forest. First, we design the adaptive feature selec-tion strategy, which can select the most distinctive ones from large number of features to separate the foreground and background.Then, we introduce the random forest classifier added with hierarchy random blocks and inspect different type of labels quickly andaccurately. Experimental results demonstrate the high accuracy and speed of our algorithm, which is suitable for high-speed produc-tion lines.

关 键 词：随机森林 自适应特征选择 标签检测 金字塔加速 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]