基于移动终端的黄瓜病害智能识别研究与应用

发布时间：2018-03-09 21:29

  本文选题：机器视觉　切入点：黄瓜叶片病害　出处：《长江大学》2016年硕士论文　论文类型：学位论文

【摘要】：新时期提出的智慧农业让农业和现代信息技术有了更密切的结合,达到的目标就是实现农业科技信息智能化、农业生产经营智能化和农业生活智能化。农业物联网是智慧农业的一种具体体现,而机器视觉技术又让农业物联网更具智能化。黄瓜是我国蔬菜种植中的主要经济作物,对病害进行准确的识别能做到对病害的预测预报及预防,对促进地方的经济发展有着重要作用。本文借助农业物联网和机器视觉技术理论基础,以识别黄瓜常见的霜霉病、白粉病、褐斑病和炭疽病四种病害为目的,利用图像处理技术为主导,根据病害叶片的颜色特征、选择合适的模式识别技术,研究了以Android智能手机为代表的移动终端通过远程诊断来自动识别黄瓜病害,识别效果很好,为黄瓜种植菜农提供了一种智能、快捷又便利地判断病害的新途径。主要研究内容和成果如下：(1)借助物联网的三层架构和机器视觉技术的基本思想,确定了本课题的实现思路。本文以Android智能手机作为图像采集设备,担当信息感知层,作为客户端。网络采用现在技术成熟的局域无线网络或移动4G网络,确保信息的上下行可靠传输,担当网络层；由于智能手机对图像处理的各函数库支持不完善,加上计算能力与PC机相比还有很大差距,图像的处理、病害特征库的建立及病害的模式识别就交由PC机来实现,PC机就是远程连接的服务器端,承担应用层的各项任务,并把识别处理的结果反馈手机客户端。(2) Android智能手机作为客户端完成的主要功能是实现病害图像的采集、存储、裁剪和联网上传及对结果接受进行显示。图像的采集调用智能手机自身的高清摄像头,采集图像存储于本地磁盘；调用本地磁盘的病害图像,定位病害突出部分进行裁剪,裁剪可以减少远程服务器的计算量,同时也简化了图像的增强和去噪声等图像预处理,让一张清晰的病害图像上传到服务器端。在联网上传的过程中,使用4G网络,采用http协议连接远程Tomcat服务器环境下的web服务器端,服务器端采用Struts2框架技术,很好地处理了手机客户端到服务器端的数据传输问题。在Android+Struts2技术的结合下,实现了图像数据快速无损传输。图像经过远程服务器端的处理和对比识别,最后把识别结果返回给手机端显示。(3)构建了一套完整的图像处理流程,快捷又成功的实现图像的病斑分割。选择红色分量灰度图像进行图像灰度化,得到病斑和背景对比清晰的灰度图,完成图像的预处理；选择一维最大熵分割法实现灰度图像的二值化,完成了背景和病斑的图像分割处理。并用图像数学形态学算法处理去掉了分割图像中的干扰杂点,完善和提升了分割效果,得到了和原彩图尺寸大小一致的,病斑和背景分离的二值图像。(4)充分研究病斑的特征,构建了病斑的颜色模型,建立了病害特征参数库。利用颜色的不同来达到识别区分不同事物是机器视觉中最为常见的一种方式。分割后的二值图像是原彩图经过灰度化和域值化处理而来,二值图像中的白色区域代表病斑区域,正常绿色背景区就对应其黑色区域。采用位置对应法,利用颜色直方图统计得到彩图病斑区域的R、G、B三分量的均值。经过实验分析发现R、G、B三分量的值会根据光照的强度变化呈正线性变化,大脑感觉色光的色度由R、G和B三分量之间的相互比值来决定,本文选择R均值和B均值分别与G均值的比作为选定的颜色特征参数。采用嵌入式、免安装、轻便型的SQLite数据库来存储各病害的特征参数。根据各病害的发病时期不同,病害的颜色表征也不尽相同,采集了各病害的早中期病害叶片图像样本多份取特征参数平均值,最后建立了每一病害早中期的病害特征库。(5)选择了适合本课题研究的模式识别方法来识别各病害。根据建立的颜色特征参数库,比较了各种模式识别方法,从中选择多类别的模板匹配模式识别方法。对四种病害各取30份,利用软件识别与普通农户及农技人员识别对比来分析,优势是明显的,识别正确率达到了91.7%。基于移动终端的黄瓜病害智能识别系统的研究,实现了实时、快速、便捷、准确、无损地进行黄瓜叶片病害诊断,解决了传统人工目测的误差和错误及农技人员的缺乏,节约了劳动力,减少了人为的主观因素,对病害做到了早知道早预防早处理,减少经济损失。方法新颖、应用潜力大、对精准农业和智能化农业的发展有着重要的意义。
[Abstract]:The new era of wisdom agriculture agricultural and modern information technology are combined more closely, to achieve the goal is to achieve intelligent agricultural information, agricultural production and operation of intelligent agriculture and intelligent life. Agricultural IOT is a concrete manifestation of the wisdom of agriculture, and the technology of machine vision and make agricultural things more intelligent. Cucumber is the main economic crops planting vegetables in our country, the disease can do for accurate identification of the prediction and prevention of the diseases and plays an important role in promoting regional economic development. With the help of agricultural IOT and machine vision technology theory, downy mildew, cucumber powdery mildew to identify common. Brown spot and anthracnose of four kinds of diseases for the purpose of using image processing technology as the leading factor, according to the color feature of diseased leaf, choose appropriate pattern recognition technology, research on Android wisdom Can the mobile phone as the representative of the remote diagnosis to automatic recognition of cucumber disease recognition, the effect is very good, provides an intelligent Cucumber Planting Vegetable, fast and convenient way to judge the new disease. The main research contents and results are as follows: (1) the basic idea with the networking of three layer architecture and machine vision technology the determination of the realization of the idea of this topic. Based on the Android intelligent mobile phone as the image acquisition equipment, as the information perception layer, network technology as a client. With the now mature wireless network or mobile 4G network, to ensure the reliable information transmission on the downlink, as the network layer; the intelligent mobile phone on the image processing function library support is not perfect, and the computing power and PC still have a large gap, image processing, pattern recognition and the establishment of disease disease feature library will be handed over to the PC machine, PC machine Is a remote connection to the server for each task in the application layer, and the recognition result of mobile phone client. (2) the main function of Android intelligent mobile phone as the client is the completion of the implementation of disease image acquisition, storage, networking and upload clipping and to display the results. The image acquisition of intelligent mobile phone calls itself HD camera, capture images stored in the local disk; call the local disk disease image, positioning disease projection clipping, clipping can reduce the amount of calculation of the remote server, but also simplifies the image enhancement and the noise of image preprocessing, make a clear disease image uploaded to a server in the process. The network upload, use the 4G network, using HTTP protocol to connect to the remote Tomcat server under the web server, the server adopts the Struts2 frame technology, very good To deal with the problem of data transmission in mobile phone client to server. In combination with Android+Struts2 technology, realizes fast lossless transmission of image data. After image processing and recognition compared to the remote server, and finally the identification results are returned to the mobile phone terminal display. (3) to construct a complete set of image processing, image the lesion is quick and successful segmentation. Choose the red component of image grayscale, contrast clear gray spots and background, complete image preprocessing; binarization choice of one-dimensional maximum entropy segmentation method to realize gray image, complete the image background and lesion segmentation with mathematical morphology and image. To remove the interference of image segmentation algorithm in noise, and improve the segmentation results, and get the original pictures of the same size, lesion and background from the value of two Image. (4) features fully study the lesion, constructed the lesion color model, established the disease characteristic parameter library. Use different colors to achieve the recognition of different things is a common way for the machine vision. After the partition of the two value image is the original image by gray scale and domain value to handle, two values of white areas represent the lesion area in the image, the normal green background area corresponding to the black area. The corresponding position method, using color histogram statistical color lesion regions R, G, B means the three component. After the experimental analysis showed that R, G, B three component values according to the there was a positive linear change of light intensity changes, your brain feels light color by R, their ratio between G and B three components to determine the choice of R mean and B mean respectively with mean G ratio as the color feature parameters selected. Using the embedded, free. The characteristic parameters of each disease, storage of portable SQLite database. According to the different period of onset of disease, disease of color characterization are not the same, the acquisition of early and mid disease disease leaf image samples from multiple copies of characteristic parameters of mean value, finally the disease feature library as early as mid every disease. (5) the pattern recognition method for the research to identify the disease. According to the color feature parameters database, comparison of various methods of pattern recognition, multi category template from matching pattern recognition method. Four kinds of diseases from each of 30 copies, analyzed by the software identification and ordinary farmers and agrotechnicians recognition comparison superiority is obvious, the correct recognition rate of 91.7%. on cucumber disease recognition system based on mobile terminal, real-time, rapid, convenient, accurate, nondestructive of Cucumber Leaves Disease diagnosis, to solve the lack of traditional artificial visual error and error and agrotechnicians, save labor, reduce the subjective factor, the disease did know early prevention and early treatment, reduce the economic loss. The novel method and application potential, is of great significance to the development of precision agriculture and intelligent agriculture.

【学位授予单位】：长江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
，

本文编号：1590299