小麦腥黑穗病害的模式识别诊断方法研究

Research on Pattern Recognition Diagnosis of Tilletia Diseases

针对小麦腥黑穗病害的形状和纹理特征选取了最有利于病害分类的6个特征参数,并分别利用最小距离法、BP神经网络和支持向量机3种模式识别方法实现了对小麦腥黑穗病害的诊断研究。对包括小麦网腥黑穗病、小麦印度腥黑穗病以及小麦矮腥黑穗病共48个孢子病害图像进行了分类诊断实验。实验表明,支持向量机法对小麦腥黑穗病的分类识别能力优于最小距离法和BP神经网络,当核函数是Sigmoid时,支持向量机法性能最优,总体识别率达到93.9%。

6 typical features are chosen from the shape and texture features of tilletia diseases,which parameters are beneficial to classify the diseases.Then three pattern recognition methods,minimum distance classifier,BP neural network and support vector machine,are used for the classification and diagnosis of tilletia diseases.48 spore images of tilletia diseases contain tilletia caries(DC.)Tul.,tilletia indica mitra,tilletia controversa Kühn are tested for the classification.The result shows that support vector machine is superior to minimum distance classifier and BP neural network.SVM responses the optimal performance with sigmoid kernel function,and the total accuracy rate reaches up to 93.9%.