基于图像识别的小麦腥黑穗病害特征提取与分类

小麦的网腥、印度腥与矮腥黑穗病危害小麦生产与人体健康,是出入境检验检疫的重要对象。该文利用小麦腥黑穗病害显微图像,采用图像分析与识别技术进行了小麦的网腥、印度腥及矮腥3类病害的分类识别。在分离出单个病害孢子图像的基础上,提取了3类病害孢子图像的16个形状和纹理特征,通过分析,从中选择小麦病害孢子的6个典型特征,并分别用最小距离法、BP神经网络和支持向量机分类器对提取的96个小麦腥黑穗病害孢子图像进行了分类试验,结果表明: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,总体识别率达到82.9%。因此,采用图像分析技术和支持向量机识别方法进行小麦腥黑穗病害诊断的方法具有可行性。

Feature extraction and classification of Tilletia diseases based on image recognition

The identification of three types of diseases of Tilletia caries (DC.) Tul., Tilletia indica Mitra and Tilletia controversa Kühn are important in the imports and exports inspection and quarantine for their harm to wheat production and human health. Three diseases were recognized and classified based on image analysis and pattern recognition techniques by using Tilletia diseases micrographs. Six typical patterns in sixteen features of shape and texture in the images of the disease infected spores were extracted. Minimum distance method, BP neural network and support vector machine (SVM) were used for the recognition and classification of 96 samples of Tilletia diseases infected spores images. The experimental results showed that the classification performance of SVM was superior to that of minimum distance method and BP neural network, the overall recognition accuracy reached up to 82.9%. Therefore, it is practicable to recognize and classify three types of Tilletia diseases by image analysis and SVM.