【作者】 司秀丽；

【导师】 袁洪印；

【作者基本信息】 吉林农业大学 ， 农业机械化工程， 2006， 硕士

【摘要】 随着计算机硬件成本的下降、CPU运算速度的提高、计算机处理能力的增强以及图像处理与识别技术本身的专业化发展，数字图像处理与识别技术在农业上的应用越来越广泛，并将成为实现农业信息化与自动化的重要技术力量。本论文以计算机图像处理技术为重要技术手段，综合运用图像处理、植物生理学、色度学、模式识别等方面的知识，研究利用计算机图像处理技术进行黄瓜霜霉病的病害识别与诊断。只有正确识别病害，才能对症下药，减少生产损失，增加农民收入。要达到对各种病症准确无误的识别，必须利用计算机图像处理与识别技术，并研制出应用软件，以克服人的视觉系统的观察误差。 本论文研究内容包括以下四个部分： 一、病害样本图像采集 在领域专家用于进行科学研究的黄瓜生产田霜霉病实验区，进行黄瓜苗期与成株期病害样本的采集。根据病害叶片的采样要求，利用光照系统和计算机图像处理装置进行病害样本的图像采集。 二、病害图像的预处理 无论采用何种装置，采集到的图像往往不能令人满意，必须对图像进行预处理以改善图像质量。本研究首先从选择图像的背景入手，选择白色作为病害图像的背景；再通过阈值处理把叶片图像从背景中分离出来；对黄瓜病害图像进行平滑处理以去除噪声；为了有效提取黄瓜霜霉病的病斑形状，采用了微分变换来对图像进行边缘检测。为下一步图像识别打下良好基础。 三、病害的识别与诊断 综合应用颜色、纹理、病斑形状三方面的特征参数，对病害进行识别和诊断。比较几种常见的色度学系统特点，进行颜色特征参数的提取；分别利用灰度共生矩阵和小波变换法对图像的纹理特征进行研究；对病害叶片病斑部位进行形状特征参数的提取。建立识别病害叶片的识别模型，构建了模式识别分类器；引入隶属度原则识别法进行模糊识别。根据黄瓜霜霉病病斑形状呈不规则的多角形这一典型特征，对黄瓜霜霉病进行识别，取得了较好的效果。 四、计算机图像处理与识别软件系统开发与试验修证 以VC++6.0作为软件开发工具，编写图像处理程序代码，最终形成适用的图像处理与识别系统。结合黄瓜生长过程，分苗期和成株期两个阶段分别进行样本采集。按更多还原

【Abstract】 Along with descending of hardware cost and increasing of CPU speed, and the increasing ability of computer process, as well as developing of specialization of image processing and identifying technology, this technology is more and more important in the way of agricultural application, and is a vital technology of achieving agricultural information and automatization as well. The main purpose of this paper is to identify and diagnose the cucumber’s downy mildew intellectively by computer image process technology. It involves in knowledge such as image process, plant physiology, chroma, pattern identify etc. We can’t suit the remedy to the disease and reduce the losing until identify correctly. Because it isn’t enough just count on human’s visual system, we should develop such a kind of software.The main contents of this paper are as follows:1. Disease samples’ image gatheringIn the experiment section which the experts study the cucumber’s downy mildew, we collect young plant and adult plant’s disease sample. According to the request of collection, we use illumination system and computer image process equipment to collect the image of disease samples.2. Disease images’ pre-processNo matter what kind of equipment we use, the images gathered won’t satisfy us, so we should pre-process the image to improve its quality. This paper first study the image’s background, and choose white as the background of disease image; and then separate the disease image from the background by binarization method; and wipe off noise by medium method; for the sake of obtaining better spot shape, we carry out edge detection by differential calculus. We have obtained better effect.3. Disease identifying and diagnosingUse color, texture, spot shape’s character parameter to identify and diagnose the diseases. Compare some familiar chroma system characteristic, and distill the character parameter of color; use gray accretion matrix and wavelet transform separately to study the texture character; and then distill the figure character parameter of the spot. Establish the identify model which identify the disease leaf, and establish the classification; introduce subjection principle to go along the faintness identify. According to the representative character, the cucumber downy mildew spot present anomalistic polygon, we更多还原