基于改进最小二乘支持向量机的小麦病虫害遥感监测研究

发布时间：2018-01-02 09:00

本文关键词：基于改进最小二乘支持向量机的小麦病虫害遥感监测研究　出处：《安徽大学》2017年硕士论文　论文类型：学位论文

【摘要】：中国作为一个农业大国,农作物病虫害发生种类多且影响范围广,给我国粮食生产造成了巨大的损失。区域尺度上准确、及时地监测农作物病虫害的发生情况有利于高效地指导防治工作,利用遥感技术对农作物病虫害信息进行提取以及区域尺度上的作物病虫害监测已经成为了热门的研究课题。然而,如何选取合适有效的方法以及最大限度地挖掘遥感影像数据中有效的信息是研究者面临的主要问题。本文以小麦的常见病虫害—小麦白粉病和小麦蚜虫为研究对象,以小麦白粉病和小麦蚜虫在区域尺度上的监测为研究主线,分别利用Landsat-8遥感卫星影像数据和环境与灾害监测预报小卫星影像数据开展小麦病虫害遥感监测模型以及方法研究,具体研究内容和成果如下:(1)给出一种粒子群优化最小二乘支持向量机的小麦白粉病监测算法。以陕西省关中平原部分地区2014年发生的小麦白粉病为研究对象,利用Landsat-8卫星OLI和TIRS数据,提取出对小麦白粉病病情影响较大的小麦长势因子和田间环境因子共5项,包括归一化植被指数(NDVI)、比例植被指数(RVI)、绿度(GREENNESS)、湿度(WETNESS)和地表温度(LST),利用最小二乘支持向量机(LSSVM)对小麦白粉病进行监测,并用粒子群优化算法(PSO)优化模型参数,将监测结果与传统最小二乘支持向量机和支持向量机(SVM)的监测结果进行对比分析。结果表明:经过粒子群算法优化的最小二乘支持向量机模型(PSO-LSSVM)的总体监测精度达到92.8%,优于传统LSSVM的85.7%和SVM的71.4%,取得了较好的监测效果。(2)给出一种基于最小二乘孪生支持向量机的小麦蚜虫遥感监测算法。以北京市通州区和顺义区2010年发生的小麦蚜虫为研究对象,基于环境与灾害监测预报小卫星HJ-CCD和HJ-IRS数据,在区域尺度上对小麦蚜虫的发生情况进行遥感监测。在小麦蚜虫发生的关键生育期(灌浆期),提取对蚜虫病情影响较大的小麦长势因子和生境因子。通过独立样本t检验的方法并结合地面调查数据对提取的特征因子进行筛选,最终选取置信度达到0.999水平的特征因子:红波段反射率、归一化植被指数(NDVI)、绿度归一化植被指数(GNDVI)、表征土壤水分含量的垂直干旱指数(PDI)以及表征小麦生长过程中田间温度状况的地表温度(LST)作为监测模型的输入变量,最后利用最小二乘孪生支持向量机建立研究区域的小麦蚜虫监测模型,并与传统支持向量机、Fisher线性判别分析和LVQ神经网络模型的监测结果进行对比。最后的研究结果表明:最小二乘孪生支持向量机模型的总体监测精度达到86.4%,优于传统支持向量机模型(77.3%)、Fisher线性判别分析模型(77.3%)和LVQ神经网络模型(72.7%),取得了较好的监测效果。
[Abstract]:China as an agricultural country, crop pest species and wide influence, caused huge losses to China's grain production. The accurate regional scale, timely monitoring the occurrence of pests and diseases to efficiently guide the prevention and control work, the use of remote sensing technology of crop diseases and insect pests on crop diseases and pests information extraction and on a regional scale pest monitoring has become a hot research topic. However, how to select the appropriate approach and maximize the effective information of remote sensing image data is the main problem faced by the researchers. Based on the common diseases of wheat powdery mildew and wheat aphid pest - as the research object, and with wheat powdery mildew wheat aphids on regional scale monitoring as the main line, respectively, using Landsat-8 satellite remote sensing image data and small environment and disaster monitoring The satellite image data to carry out wheat diseases and pests monitoring model and research method, the main research contents and results are as follows: (1) proposed a monitoring algorithm of Wheat Powdery Mildew in particle swarm optimization least squares support vector machine. In 2014 some region of Guanzhong Plain in Shaanxi province wheat powdery mildew as the research object, using Landsat-8 OLI and TIRS satellite data. Extract of wheat powdery mildew disease affecting wheat growth factor and field environment factor of 5, including the normalized difference vegetation index (NDVI), ratio vegetation index (RVI), green (GREENNESS), relative humidity (WETNESS) and surface temperature (LST), using the least squares support vector machine (LSSVM) monitoring of wheat powdery mildew, and by using particle swarm optimization (PSO) algorithm to optimize the parameters of the model, the monitoring results and the traditional least squares support vector machine and support vector machine (SVM) compared with the monitoring results The results show that the least squares analysis. Through particle swarm optimization model of support vector machine (PSO-LSSVM) monitoring the overall accuracy of 92.8%, better than the traditional LSSVM 85.7% and SVM 71.4%, achieved a good monitoring effect. (2) proposed a least squares twin support vector machine algorithm based on remote sensing monitoring of wheat aphids. Beijing City, Tongzhou District and Shunyi District in 2010 occurred in wheat aphids as the research object, the environment and disaster monitoring satellites HJ-CCD and HJ-IRS based on the data of occurrence of wheat aphids in regional scale remote sensing monitoring. In the key growth period of wheat aphids (the filling stage), extraction of large wheat growth factor and environmental factor effect on aphid disease. Through independent sample t test method and combined with the characteristic factor of the ground survey data on the extraction were screened, the final selection of confidence up to 0.999 water Eigenfactor flat: red band reflectance, normalized difference vegetation index (NDVI), green vegetation index (GNDVI), perpendicular drought index to characterize soil moisture content (PDI) and the growth process of wheat field temperature characterization of land surface temperature (LST) measurement model as input variables of supervision, the monitoring of wheat aphids the establishment of regional model of least squares twin support vector machine, and the traditional support vector machine, comparative analysis of monitoring results and LVQ neural network model Fisher linear discriminant. The final results show that: the overall precision of least squares twin support vector machine model reached 86.4%, better than the traditional support vector machine (77.3%), Fisher linear model (77.3%) the discriminant analysis model and LVQ neural network model (72.7%), good monitoring effect was obtained.

【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S435.12;S127;TP18

【参考文献】