4.2纹理特征提取-Read.doc

基于图像分析的苹果病害识别技术研究 摘 要 本文以手机拍摄图像分析进行苹果病虫害识别为目标，以苹果叶部病害为对象，以发生面积较大、危害较为严重的苹果斑点落叶病、锈病、圆斑病、花叶病、黄叶病等5种病害为重点，研究基于图像分析的苹果早期病害的识别与分类。主要研究内容及结论如下： （1）针对自然条件下用手机拍摄的低分辨图像噪声类型的复杂性，深入研究图像预处理中的去噪问题，建立了完整的预处理流程：对图像用三段线性法灰度调整，扩展图像灰度的动态范围，用形态学Top-Hat变换消除光照；对反色运算后的彩色图像进行RGB通道分离，并分别用改进的加权中值滤波法去噪，然后进行通道融合，得到增强图像；最后对增强后的图像RGB分量分别用改进的类间方差阈值分割法、形态学运算、区域标记等方法抽取病叶的病态部位。结果表明，红色（R）分量分割效果最为理想，可用于对病斑区域的正确分割。 （2）研究了苹果病害图像的特征提取，分别从颜色特征、纹理特征和形状特征3个方面对测试图像进行实验，即根据病斑几何特征及Hu不变矩提取病斑的形状特征，提取病斑的H方差结合H-S直方图特征作为病斑的颜色特征，用计盒维数法提取病斑纹理特征，取得了较好的效果。从试验中的27个特征优选出8类作为分类特征参数。 （3）分析了4类模式识别方法，选用并构建BP神经网络病害识别模型。对苹果5种病害图像自动识别实验结果表明，平均正确率达92.6%，能够满足苹果远程诊断的需要。 （4）运用VC++和Matlab混合编程，实现了基于手机拍摄图像分析的苹果病害识别系统相关功能模块的设计，为广大农技人员和农民提供方便快捷的苹果安全生产、主要病害识别与治理及合理用药等技术咨询服务平台。 论文研究结论对于促进计算机图像分析技术在其他农作物病虫害远程诊断的研究与应用具有一定的借鉴意义。 关键词：苹果；病害识别；图像分析；图像特征；神经网络 RESEARCH ON IDENTIFY TECHNOLOGIES OF APPLES DISEASE ABSTRACT This paper used the images of apple pests which captured by mobile phone to identify the pests and diseases of apple, and the emphasis on apple leaf involved five kinds of diseases such as spot, rust, round spot, mosaic, yellow leaf disease, etc. We specially focused on the diseases whose place is larger and more serious damaged. And all of the research based on image analysis of Apples early disease identification and classification. Main contents and conclusions are as follows: (1)Focusing on the complex type of noise in low-resolution images which is shot under natural conditions using a mobile phone, and in-depth study of the image pre-processing of the denoising problem, we set up a complete pre-treatment process: The image sections of the linear method with three gray adjust, expanded the dynamic range of gray, used Top-Hat transform to eliminate the effect of light; Treated the color image with the anti-color operation, and enhanced image through the RGB channel separation, median filtering, channels fusion; Finally, we used improved between-class variance threshold method and morphological operations and region labell