【作者】 王健；

【导师】 王波涛；

【作者基本信息】 北京工业大学 ， 电子与通信工程（专业学位）， 2017， 硕士

【摘要】 目前针对农业生产环节的基础设施和信息化服务水平显著滞后。表现为:农业环境要素、作物长势状况、灾害发生情况等无法实时监测;决策管理部门、各级管理人员、一线生产者和科学家等难以实现信息实时共享。因此,本文基于国内外农业信息化服务体系目前的发展概况,针对国内农业信息化建设存在的不足之处,在ASP.NET平台下,基于SQL Server数据库,网络通信,微处理器,流媒体通信等技术,并结合图像模式识别理论,研究并开发了一种可视化病虫害信息采集与处理系统。本文主要研究内容包括以下几个方面:1定制可视化病虫害信息采集与处理系统总体方案,设计并组建采集系统硬件平台,并分别实现了能够实时采集病虫害信息的硬件系统片上程序软件和数据存储软件。2针对孢子病菌显微图像粘连、噪声大导致的难分割、过分割问题。以小麦白粉孢子为例,分析了影响分割准确度的主要因素。提出一种先以高低帽变换对图像进行增强,再以中值滤波器去除孤点噪声,并用改进的形态学滤波器处理距离变换前的二值图像,再对距离变换的灰度图像进行拓展极小值和形态学重建,最后应用分水岭分割算法的方法,实现了孢子图像的准确分割。并依据6种孢子图像的特征筛选出待识别的感兴趣区域,去掉了大部分伪区域,减少了识别判断的数据量。3选取HOG特征,LBP特征,Haar特征,分别应用AdaBoost分类器进行训练。并对分割出的感兴趣区域进行识别。对比分析后得出,基于LBP特征的分类器检测效果最好。4在ASP.NET平台下,实现了基于B/S架构上位机系统软件,完成了整个系统的开发,并对软件进行了实地测试,对其测试结果进行了分析,实际软件识别测试平均准确率为76%,平均误检率为19%。更多还原

【Abstract】 At present,the level of infrastructure and information service for agricultural production is lagging behind.The performance of agricultural environmental factors,crop growth situation,the occurrence of disasters can not be real-time monitoring;Decision-making management and front-line producers and scientists are difficult to achieve real-time information sharing.Therefore,based on the current development of agricultural information service system at home and abroad,this paper aims at the shortcomings of domestic agricultural information construction,under the ASP.NET platform,based on SQL Server database,network communication,microprocessor,streaming media communication Technology,combined with the image pattern recognition theory,we research and develop a visualization of pest and disease information collection and processing system.The main research contents include the following aspects:1.We identify a visualization of pest and disease information collection and processing system solution,designe and set up the acquisition system hardware platform,and realize real-time acquisition of pest and disease information hardware on-chip software and data storage software program.2.Aiming at the problem of microscopic image adhesion and sporulation of spore pathogens,it is difficult to divide and overdrive.We take the wheat flour powder as an example,analyzed the main factors affecting the accuracy of segmentation.A new method is proposed to enhance the image by using the high and low cap transform,then the residual filter is removed by the median filter,the binary image before the distance transformation is processed by the improved morphological filter,and the gray scale image of the distance transformation is extended Minimization and morphological reconstruction.Finally,the method of watershed segmentation is applied to realize the accurate segmentation of spore image.According to the characteristics of six kinds of spore images to identify the region of interest to be identified,removed most of the pseudo-area,reducing the amount of identification to determine the ROI regions.3.We experimentally select HOG characteristics,LBP features,Haar characteristics,respectively,using AdaBoost classifier for training.Which will be used in the region of interest identification.Compared with the analysis,it is concluded that the classifier based on LBP is the best.4.In the ASP.NET platform,we achieved based on the B / S architecture PC system software,completed the entire system development.The software has been field tested and analyzed the test results.The average accuracy rate of the actual software identification test is 76% and the average false alarm rate is 19%.更多还原