目录
摘 要 IV
关键词 IV
Abstract IV
引言
1 综述 1
1.1 研究背景 1
1.2 研究目标 2
1.3 研究内容 2
1.4 开发工具简介 2
2 系统分析 3
2.1 可行性分析 3
2.2 需求分析 3
2.2.1用户需求分析 3
2.2.2 系统需求分析 4
3 系统设计 5
3.1 概要设计 5
3.1.1 功能模块设计 5
3.1.2各模块功能及关系 5
3.2 详细设计 5
3.2.1 作物病害图谱模块设计 5
3.2.2 作物病害检测模块设计 6
3.2.3系统类图 7
3.2.4 数据库设计 8
4 系统实现 9
4.1 开发环境 9
4.2 病害图像采集 9
4.3 图像预处理 9
4.3.1 图像增强 9
4.3.2 图像去噪 10
4.4 图像分割 10
4.4.1 基于阈值的图像分割 10
4.4.2 基于边缘的图像分割 12
4.5 特征提取 14
4.5.1 Hu矩 14
4.5.2 LBP特征 15
4.5.3 颜色特征 18
4.5.4 Hog特征 20
4.6 图像分类 21
4.6.1 巴氏距离 21
4.6.2 支持向量机 21
4.7 前端设计 23
5 系统测试 25
5.1基于Hu矩的巴氏距离识别方式 25
5.2 基于LBP特征的SVM识别方式 25
5.3 基于颜色特征的SVM识别方式 25
5.4 基于Hog特征的SVM识别方式 26
5.5 结论 26
致谢 26
参考文献 26
 *51今日免费论文网|www.51jrft.com +Q: ^351916072# 
作物叶部典型病害检测系统
摘 要
传统的作物病害诊断主要通过人工方式，需要种植者拥有一定的经验和相关知识，并且由于人主观意识的不同，容易出现误判情况。另外随着现代农业的发展，种植规模逐步扩大，使用传统的人工方式对作物进行病害检测的工作量过大，效率低下。针对这一现象，基于计算机图像处理技术对作物叶部病害进行无损检测，可提高效率。系统以马铃薯和苹果的五种病害为研究对象，基于PlantVillage收集数据，以Mysql数据库存储数据。首先对图像进行直方图均衡化及灰度变换增强、中值滤波去噪、最大类间方差法分割等预处理；然后提取多种典型特征，主要有Hu矩特征、颜色矩特征、LBP局部二值模式特征和Hog梯度方向特征；最后使用支持向量机SVM基于特征训练样本并分类。对比分析了基于以上特征检测病害的综合性能，发现基于Hog特征的SVM分类算法性能最优，故在系统实现时采用了此种方法。系统采用B/S模式，前端使用Java Swing进行界面设计，后端基于OpenCV提供的接口实现对图像的处理。论文最后对系统进行了功能性测试，基本达到了设计目标。
TYPICAL DISEASE DETECTION SYSTEM FOR CROP LEAVES
ABSTRACT
The traditional crop disease diagnosis is mainly through manual methods, which requires the grower to have certain experience and related knowledge, and due to the different subjective consciousness of the person, it is prone to misjudgment. In addition, with the development of modern agriculture, the scale of planting is gradually expanding, and the workload of using traditional manual methods to detect crops is too large and the efficiency is low. In response to this phenomenon, nondestructive detection of crop leaf diseases based on computer image processing technology can improve efficiency. The system takes five diseases of potato and apple as research objects, collects data based on PlantVillage, and stores data in Mysql database. First, preprocess the image with histogram equalization and grayscale transformation enhancement, median filter denoising, and maximum interclass variance method segmentation; then extract a variety of typical features, mainly Hu moment feature, color moment feature, local binary pattern features and Hog features; finally, support vector machine is used to train and classify the samples based on the features. By comparing and analyzing the comprehensive performance of disease detection based on the above features, it is found that the SVM classification algorithm based on Hog features has the best performance,so this method is used in the system implementation. The system adopts B/S mode, the front end uses Java Swing for interface design, and the back end implements image processing based on the interface provided by OpenCV. At the end of the thesis, the system was functionally tested and basically reached the design goal.

原文链接：http://www.jxszl.com/jsj/wljs/606872.html