摘要： 介绍了采用三帧差分法实现谷物害虫图像恢复与提取的方法，利用图像的一阶灰度值直方图和图像的目标区域，自动提取静态储粮害虫图像的纹理等特征。针对于相对特征维数而言样本数很少的特点，提出利用多类SVM 分类器的方法实现对储粮害虫的快速鉴定和分类。实验结果表明，相比传统的神经网络，SVM 在有限样本情况下具有良好的泛化能力。

关键词: 三帧差分法；纹理特征；分类；人工神经网络；SVM

Abstract: A method of using three image differences to restore and extract the pest images is introduced. With reference to the fist order grayhistogram of pest images and their graphic target zones, a technique is provided to extract the texture eigenvalue. A multi-class support vector machine is proposed to implement the quick identification and classification of grain pests according to small amount samples compared to the dimensions. The test shows the better generalization ability of SVM than that of ANN under the conditions of limited training samples.

Key words: Three image difference; Texture eigenvalue; Classification; ANN; SVM