基于病斑特征融合的烟草病害图像检索方法

Image Searching Method of Tobacco Disease Based on Disease Spot Feature Fusion

针对目前烟草病害诊断专家系统依靠肉眼获取病害特征,致使病害诊断存在不确定性、误判等现象,提出了一种基于病斑特征融合的烟草病害图像检索方法以诊断病害。通过图像处理技术分割病害图像病斑,提取病斑的颜色、纹理、形状特征,依据双编码遗传算法和支持向量机识别病害模型对病斑特征降维,以获取表征病害的有效特征及权重。将有效特征归一化处理后与病害图像数据库系统中的病斑特征进行图像相似度距离计算,按距离大小返回一批相似图像,依据相似图像获取病害描述及防治措施。以烟草7种常见病害进行试验表明,融合病斑的颜色、纹理、形状特征检索病害图像,查准率和查全率明显比用单个特征检索的高。用这种方式诊断烟草病害,不但有较高的病害识别率,还能使诊断结果可视化,将其用于烟草病害诊断专家系统,将提高系统的鲁棒性,为实现病害的远程在线诊断提供条件。

Recently,tobacco disease diagnosis expert system relies on the naked eye to diagnose disease characteristics,which causes disease diagnostic uncertainty and misjudgment. This study proposes a disease spot image searching method to diagnose tobacco diseases based on disease spot feature fusion.The color,texture and shape features of a disease are extracted after disease spot image segmentation,and then disease spot feature dimension reduction is conducted based on double coding genetic algorithm and support vector machine（ SVM） model to obtain effective features for characterization of the disease characteristics and respective weights. The image similarity distance calculation is processed between the normalization of effective characteristics and the disease spot features of the image database system. According to the distance a number of similar images are returned,by which the descriptions of diseases and preventing measures are presented. Experiments with seven common diseases of tobacco showed that the precision and comprehensiveness of disease image retrieval based on fusion of color,texture and shape features were significantly higher than those using a single feature retrieval. Diagnosis of tobacco disease in this way,not only has higher recognition rate of diseases,but also realizes visualization of diagnosis results. Application of this method in tobacco disease diagnosis expert system will improve the robustness of the system,providing conditions for realizing remote online diagnosis of diseases.