基于属性降维的粮库害虫检测与识别  

Grain Pests Detection and Classification Based on Attribute Reduction

作　　者：范艳峰[1,2] 吴建军[1] 甄彤[1] 

机构地区：[1]河南工业大学信息科学与工程学院 [2]西北工业大学计算机学院,陕西西安710072

出　　处：《微电子学与计算机》2006年第11期12-15,共4页Microelectronics & Computer

基　　金：国家"粮食丰产科技工程"项目(2004BA520A)

摘　　要：介绍了害虫检测系统的组成,采用三帧差分法实现害虫图像的恢复与提取,利用图像的一阶灰度值直方图和图像的目标区域,自动提取静态害虫图像的纹理等特征。通过条件属性的可分性度量值作为判断条件属性重要性的方法,利用神经网络进行属性选择。实验结果表明,该方法简化了神经网络的结构,提高了分类效率。The composition of the pest detection system was introduced. Three image difference was adopted to restore and extract the pest images. With reference to the fist order gray histogram of pest images and their graphic target zones, a technique was provided to extract the texture eigenvalue. Separability criterion was used to rank the attributes and attributes were selected by using neural network. As experiments demonstrated, this method simplified the structure of the neural network and improved the efficiency of identification.

关 键 词：三帧差分法 纹理等特征 可分性度量 属性选择 人工神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]