基于图像处理的酿酒葡萄叶片病害分割与诊断  

Study on segmentation and diagnosis of wine grape disease based on image processing

作　　者：朱林[1] 冯全[2] 杨梅[2] 张子鹏[3] 

机构地区：[1]兰州资源环境职业技术学院机电工程系,兰州市730021 [2]甘肃农业大学工学院,兰州市730070 [3]华北电力大学控制与计算机工程学院,北京市102206

出　　处：《中国农机化学报》2015年第1期111-115,共5页Journal of Chinese Agricultural Mechanization

基　　金：盛彤笙科技创新基金项目(GSAU-STS-1324);国家自然基金资助项目(61062012)

摘　　要：利用图像处理技术和人工神经网络对酿酒葡萄常见的4种病害进行分割和识别。首先,在HSV颜色空间中采用ISODATA聚类算法分割病斑,在病斑区域,对于H、S、V分量分别提取了基于灰度共生矩阵的4种纹理特征以及颜色特征;然后以人工神经网络为分类器,对各分量及其组合的特征对于识别精度的影响进行了实验研究,结果表明,H分量对于4种病斑均具有较好的分割效果;在诊断时,HS组合的特征具有最好的识别效果,平均准确率达到了90%,对白粉病识别率则达到94%。The segmentation and recognition methods of four kinds of wine grape diseases were studied by image processing technique and artificial neural nets. Firstly, an unsupervised method based on ISODATA cluster algorithm was proposed to segment disease region in HSV color space. For each component of HSV, the four texture features derived from gray-level co-occurrence matrix and the color feature were extracted. Then classifiers of artificial neural net were trained to identify the type of diseases utilizing the features of each component and their combinations. The best segmentation result was achieved by H component, and the best recognition result was achieved by the features of HS combination, which average accuracy is 90% for all kinds of diseases and 94% for powdery mildew.

关 键 词：酿酒葡萄 病害诊断 ISODATA 图像分割 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]