【作者】 吴艳兵；

【导师】 樊啟洲；

【作者基本信息】 华中农业大学 ， 农业机械化工程， 2007， 硕士

【摘要】 在工厂化农业高速发展的今天，作物栽培管理的自动化和现代化日趋重要。应用现代无损伤测量和检测手段进行作物长势诊断的技术研究，成为现代农业技术发展的热点问题之一。本研究选择温室黄瓜作为研究对象，在自然光条件下进行试验，运用计算机图像处理技术，对黄瓜的叶片图像特征进行提取，分析黄瓜植株的生长信息，进而实现对土壤水份含量的检测，检测结果可作为精准灌溉的参考，为后期的灌溉提供科学的依据。基于以上目的，本研究主要进行了以下工作：(1)按照常规方法对植株的生长进行管理，试验中土壤水份含量设计5个处理水平：90％、80％、70％、60％和50％，浇水量按照土壤水份含量各处理水平的±5％控制。使用数码相机统一采集自植株顶部下数第3片功能叶的图像。(2)对叶片图像进行直方图修正、平滑和锐化等预处理，再分割出叶片和背景。提出了一种适合于自然光条件下分割叶片和复杂背景的方法，采用过绿分割法对图像进行分割，并运用自动选择分类阈值的最大方差比法确定分割阈值T，该方法能很好的将叶片和背景进行分割。(3)运用图像处理技术，提取出叶片图像各颜色特征值。通过分析土壤水份含量与各颜色特征的线性相关关系得出，土壤水份含量分别与颜色特征r、H之间均呈高度相关性，并且达到了显著性检验水平(p=0.05)，同时，它们的值域变化区间和土壤水份含量水平之间存在有较好的对应关系。选用颜色特征参数r分量和H分量作为检测系统的检测指标，对土壤水份含量进行检测是可行的。(4)在检测系统的检测指标和土壤水份含量之间建立三层BP神经网络，采用自适应学习速率梯度下降反向传播算法对网络进行训练，训练后的网络对训练样本的正确识别率为100％，对测试样本，处理水平50％至90％的识别率分别为92.5％、97.5％、100％、97.5％和100％。以VC++为开发平台，编制出土壤水份含量的检测系统。本研究对于计算机图像处理技术应用于温室作物实际生产，并最终提高工厂化农业的智能化管理水平具有重要的意义。更多还原

【Abstract】 As industrialized agriculture technology progress rapidly, the automatization for cropraising is becoming more important. The nondestructive measurement and testing is avery hot agricultural production technology. Choosing greenhouse cucumber as theresearch object, cucumber leaves image feature was obtained using computer imageprocessing technology under natural illumination, and the growth information ofcucumber was analyzed, then realizing the moisture content of soil was tested, theresults could be used as a reference for precision irrigation, and provide scientific basisfor the latter part of the irrigation.Based on above goals, the following work was finished in this study:(1) Managing the plant according to conventional methods, the soil moisture contentwas set as five levels: 90%、80%、70%、60% and 50%. The amount of water wascontrolled with the±5% of soil moisture content. The images of the No.3 leaf wereacquired using digital camera.(2) The leaf images were used to pass through histogram correcting, smoothing andsharpening, and then segmented they were leaves and background. A suitable method,super-green, for segmenting leaves and complex background, was advanced undernatural illumination, and the maximal variance ratio subtract method which could chooseautomatically threshold was used to compute the threshold segmentation T, which waseffective by experiment.(3) Using computer image processing technology, color feature of leaf image wasobtained. There are both a high correlation between soil moisture content and colorfeatures r, H, by analyzing the linear correlation between soil moisture content and colorfeatures, which achieved confidence level (p=0.05), and then, there is a well relationshipbetween the range of r, H and soil moisture content. It is feasible that color feature r, Hwere selected as testing parameters of the testing system.(4) The three tiers BP neural network was designed between the testing parametersand soil moisture content. The network was trained by a reverse transmission algorithmof self adaptive learning rate with gradient reduction, after training the network, thecorrect recognition rate is 100% for training samples, it is 92.5%, 97.5%, 100%, 97.5%and 100% for treatment level 50% to 90% of the tested samples. The testing system wasprogrammed with the development platform of VC++.This study may provide a significant help to the greenhouse crop production bycomputer image processing technology, and ultimately to improve the intelligentmanagement level of industrialized agriculture.更多还原