基于过渡区研究的黄瓜病害识别方法

Cucumber disease identification method based on transition area research

【目的】针对黄瓜生长过程中常见的霜霉病、白粉病和靶斑病,提出改进的病害识别方法,为黄瓜病害自动识别提供一种技术支持。【方法】将RGB模型的病害图像转换到HSV和YUV颜色空间,通过OTSU筛选,获取阈值分割效果最好的HSV颜色空间的V分量,综合全阈值法和局部动态阈值法对V分量进行分割,获取病斑区和过渡区的分割图像。分别提取病斑区和过渡区的颜色和形状特征,基于支持向量机(Support vector machines,SVM)进行病害识别。【结果】以采集的240幅病害图像为研究样本,当惩罚参数C=32,核函数参数γ=1时,基于病斑区和过渡区在颜色和形状方面的22个特征数据,SVM分类器对霜霉病、白粉病和靶斑病3种病害的识别率分别达83.3%、76.7%和90.0%,对比仅以病斑区的11个特征数据为基础的识别结果,增加过渡区特征数据之后,黄瓜病害识别率有较大提升,分别提高26.6%、13.4%和16.7%(绝对值)。【建议】未来研究中应拓展黄瓜病害研究的种类,在进行病害识别时应将病害发展程度及病害的混合性考虑在内。

【Objective】To improve the disease identification method for common downy mildew,powdery mildew and target spot disease in cucumber growth process,and provide a technical support for automatic identification of fruit and vegetable diseases.【Method】The disease image of RGB model was converted to HSV and YUV color space. The V component of HSV color space with the best threshold segmentation effect was obtained by OTSU. The V component was segmented by integrated full threshold method and local dynamic threshold method and obtained a segmented image of the lesion area and the transition area. The color and shape features of the lesion area and the transition area were extracted, and the disease recognition was performed based on the support vector machine(SVM).【Result】Taking 240 disease images as research samples,when penalty parameter C=32 and kernel function parameter γ=1,based on the 22 features of the lesion area and the transition area in color and shape,the recognition rates of the SVM classifier for downy mildew,powdery mildew and target spot disease reached 83.3%,76.7% and 90.0%. Compared with the identification results based on 11 features of the lesion area,after increasing the features of the transition area,the disease recognition rate greatly improved 26.6%,13.4% and 16.7%(abslute value).【Suggestion】In the future,the research should expand the types of cucumber disease research,and take into account the degree of disease development and the mixture of diseases in the identification of diseases.