黄土高原苹果叶面病害图像识别方法研究

发布时间：2018-05-07 03:14

本文选题：苹果 + 病害识别　；参考：《西北农林科技大学》2017年硕士论文

【摘要】：苹果产业对陕西和甘肃等地区的经济有着重要的影响,是黄土高原地区的特色产业之一。由于得天独厚的自然条件,黄土高原地区的苹果品质十分优良。近年来苹果产业蓬勃发展,种植面积不断扩大,但是由于自然灾害和治理不当等原因,苹果病虫害频发,严重影响了苹果的产量。针对黄土高原苹果病害频发又得不到及时有效治理的问题,本文对该地区苹果叶面比较常见的病害:斑点落叶病、花叶病和锈病进行了相关的资料调查和研究,并采集了相应的图像作为样本,应用图像处理技术对图像进行处理和特征的分析,开发出苹果叶面病害识别系统,最终实现3种病害的识别。本文的主要研究内容如下:(1)针对复杂背景下的苹果叶面病害图像特点,深入研究了图像预处理中的去噪和病斑分割问题,建立了完整的预处理流程:采用三段线性法对图像进行灰度变换,扩展灰度动态范围;利用改进的中值滤波方法对图像进行滤波处理,该方法可以有效地去除噪点,增强图像信息;将图像由RGB颜色空间转化到L*a*b*颜色空间,并使用K均值聚类方法将叶面与背景分割,然后采用改进的最大类间方差法对分离出的叶面图像进行分割,得到病斑图像。实验表明,这种病斑分割方法可以达到较为理想的分割效果。(2)研究了苹果病害图像的有效特征的提取,分别从颜色特征、形状特征和纹理特征三方面对测试对象进行实验,提取病斑的H方差并绘制H-S直方图作为病斑的颜色特征,根据病斑的几何特征和Hu不变矩提取病斑的形状特征,采用灰度共生矩阵对病斑纹理进行分析,从实验中的22个特征优选出13个作为分类特征参数。(3)研究模式识别相关方法和支持向量机模型,并通过与贝叶斯(Bayes)决策方法、人工神经网络方法优缺点的比较,选用基于支持向量机的病害分类模型,根据一对一投票策略设计出多分类情况下支持向量机的分类器模型,测试并确定其模型参数,对优选出的13个特征进行分类训练。(4)本文采用C#代码做编程实验,并在C#平台中加入Matlab程序接口,开发出识别系统。实验结果表明,该分类方法能够对3种苹果叶面病害图像进行有效识别,可以满足苹果病害智能诊断的需要。
[Abstract]:Apple industry has an important impact on the economy of Shaanxi and Gansu provinces and is one of the characteristic industries in the Loess Plateau. Due to the unique natural conditions, the apple quality in the Loess Plateau is very good. In recent years, the apple industry is booming and the planting area is expanding. However, due to natural disasters and improper management, apple diseases and insect pests occur frequently, which seriously affects the production of apple. In order to solve the problem of apple diseases occurring frequently and without timely and effective treatment in the Loess Plateau, this paper investigated and studied the common diseases of apple leaves in this area: speckle disease, mosaic disease and rust disease. The corresponding images were collected as samples, and the image processing technology was used to process and analyze the characteristics of the image. The apple leaf surface disease recognition system was developed. Finally, the recognition of three kinds of diseases was realized. The main contents of this paper are as follows: (1) in view of the characteristics of apple leaf surface disease image under complex background, the problem of de-noising and disease spot segmentation in image preprocessing is deeply studied. A complete preprocessing flow is established: the image is transformed into gray scale by three-segment linear method to extend the dynamic range of gray scale, and the improved median filtering method is used to filter the image, which can effectively remove the noise. Image information is enhanced, the image is transformed from RGB color space to Lena color space, and the leaf surface and background are segmented by K-means clustering method, and then the separated leaf surface image is segmented by the improved maximum inter-class variance method. Get an image of the disease. The experimental results show that this method can achieve an ideal segmentation effect. The effective feature extraction of apple disease image is studied. The experiment is carried out from three aspects: color feature, shape feature and texture feature. The H variance of the disease spot is extracted and the H-S histogram is drawn as the color feature of the disease spot. According to the geometric feature of the disease spot and the Hu invariant moment, the shape feature of the disease spot is extracted, and the gray level co-occurrence matrix is used to analyze the texture of the disease spot. From the 22 features in the experiment, 13 are selected as classification feature parameters.) the correlation method of pattern recognition and support vector machine model are studied, and the advantages and disadvantages of artificial neural network are compared with Bayes Bayes decision method. Based on the support vector machine (SVM) based disease classification model, the classifier model of support vector machine (SVM) is designed according to one-to-one voting strategy, and its model parameters are tested and determined. In this paper, we use C # code to do programming experiment, and add Matlab program interface to C # platform to develop the recognition system. The experimental results show that the classification method can effectively identify the three apple leaf disease images and can meet the need of intelligent diagnosis of apple diseases.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S436.611;TP391.41

【参考文献】