自然场景下低分辨率苹果果实病害智能识别方法

The Intelligent Identification Methods of Apple Fruit's Disease in Nature Outdoors Based on Low-resolution Image

为实现自然场景下低分辨率苹果果实病害的智能识别,对获取图像进行预处理,采用改进的水平集交互式分割方法提取病斑。根据病斑特点,提取H,S,V等3个通道的2个低阶颜色矩作为颜色特征,基于灰度共生矩阵提取8个特征参数作为纹理特征,提取病斑的Hu不变矩作为形状特征。在对特征进行优选的基础上,构建支持向量机病害识别模型。实验结果表明,用优选的15个特征和支持向量机识别模型,对3种病害的平均正确识别率达到90%,可以有效识别苹果果实病害。

To realize intelligently identification of the apple fruit's disease in nature outdoors based on low-resolution image,after pre-processing,the diseased parts were extracted using improved level set interactive segmentation method.According to the characteristic of the disease spots,two low color moments of H,S and V channels were extracted as color characteristic,the co-occurrence matrix was used to extract eight texture characteristics and seven Hu invariant moments as shape feature.After choosed the preferred characteristics,the disease identification model is constructed based on the support vector machine classifier.Experiments show that the optimal 15 features and the support vector machine model have the average correct identification rate arrived 90% for the 3 kinds of diseases,which can effectively identify the apple diseases.