【作者】 唐俊；

【导师】 周新志；

【作者基本信息】 四川大学 ， 模式识别与智能系统， 2021， 硕士

【摘要】 监测作物生长状况,预测作物产量对国家粮食安全和农业可持续发展至关重要。MODIS数据由于其探测周期短、获取成本低及覆盖范围广等优点,常常被运用于监测地面的生产活动。传统的基于遥感数据的长势监测及产量预测模型是以市级、省级或平原等作为研究区域,在指导农业的精准化生产中,这些模型的精度往往具有一定局限性。除此之外,部分模型还需要建立特殊的农业站点或是利用到气候、温度等一些复杂的统计数据,这使得模型的推广性也不强。因此,为提升模型的精度及推广性,本文选取黑龙江省哈尔滨市宾县为研究区域,以该区域的主要作物玉米为研究对象,采用时序MODIS影像及产量统计数据作为主要数据源展开了县级尺度的作物长势监测及产量预测研究,主要研究内容与结果如下:(1)作物长势监测及EVI值预测研究。为宏观了解整个研究区域内作物生长走势及长势优劣,利用时序MOD09A1数据提取了研究区域玉米2000年-2019年的EVI(Enhanced Vegetation Index)时间序列。并通过EVI时间序列进行曲线间的对比分析完成了作物长势过程监测。同时,为了实时了解研究区域内作物的长势,基于EVI建立了RPEVI。除此之外,提出并建立了使用RBF神经网络预测植被指数的方法。通过建立预测日期以前的EVI时间序列与预测日期EVI值之间的非线性关系,完成了EVI值的预测。本文所构建的EVI-RBF模型的预测精度在90%以上,且预测曲线与实际曲线的走势一致。(2)作物产量预测研究。在完成玉米产量预测的研究中,针对传统的线性回归模型失效的问题,本文提出了利用RBF神经网络完成玉米产量的预测的方法。利用RBF神经网络建立了玉米关键生长时期的植被指数时间序列与产量间的非线性关系。分别使用线性回归模型、RBF模型以及BSSO(Bound Social Spider Optimization)-RBF模型验证了常用的植被指数EVI、NDVI(Normalized Difference Vegetation Index)、RVI(Ratio Vegetation Index)及SAVI(Soil-Adjusted Vegetation Index)在产量预测中的适用性。结果显示,BSSO-RBF模型取得了最佳预测精度以及与产量最高的相关性。同时,EVI在产量预测过程中呈现了最高的适用性。(3)BSSO算法优化RBF神经网络研究。本文在使用RBF模型预测玉米产量时,由于所获取的遥感数据仅有20年,样本集数目较小,导致RBF模型在训练中容易陷入局部最优,进而难以保障预测精度。因此,本文在SSO(Social Spider Optimization)算法基础上做出了相应修改并提出了BSSO算法,并使用BSSO算法优化了RBF神经网络。研究在三个普通数据集进行了对比实验,验证BSSO算法与SSO算法、粒子群算法(PSO)以及另一种社会蜘蛛算法——SSA(Social Spider Algorithm)的优化性能。实验结果显示,BSSO算法在所有数据集中的迭代次数、预测精度及稳定性上均表现出了更为优异的性能。本文在作物长势监测的基础上开展的长势预测研究以及在线性回归模型估产失效时开展的RBF模型估产及BSSO-RBF模型估产研究,不仅实现了能够从实时及过程两个角度监测作物长势,从而协助相关部门判断长势优劣,还能够完成植被指数及产量的高精度预测,从而帮助有关部门提前调控相关农业生产活动并提前制定高效的粮食政策,具有较好的应用价值。更多还原

【Abstract】 Monitoring the growth of crops and predicting crop yields are critical to national food security and sustainable agricultural development.MODIS data has been widely used in earth observation activities due to its advantages of short detection period,low cost and wide coverage.Traditional growth monitoring and yield prediction models based on remote sensing data always take municipal,provincial or plain areas as the research area.In guiding the precision production of agriculture,the accuracy of these models often has certain limitations.In addition,some models also need to establish special agricultural sites or use some complex statistical data such as climate and temperature,which makes the generalization ability of the model not strong.Therefore,in order to improve the accuracy and generalization of the model,this article selects Bin County,Harbin City,Heilongjiang Province as the research area,takes the corn as the research object,and takes the time series MODIS images and yield statistics data as the main data sources to carry out the research on the growth monitoring and yield prediction of maize at the county level.The main research contents and results are as follows:(1)Research on crop condition monitoring and EVI value prediction.In order to assess the trend of crop growth and the pros and cons of the growth in the whole study area from a macro perspective,based on MOD09A1 data,the time series of corn EVI(Enhanced Vegetation Index)from 2000 to 2019 in the study area is extracted.And through the comparison and analysis of the EVI time series,the monitoring of the crop growth process has been completed.At the same time,in order to monitor the growth of crops in the study area in real time,RPEVI was established based on EVI.In addition,a method of using RBF neural network to predict vegetation index is proposed and established.By establishing the non-linear relationship between the EVI time series before the forecast date and the EVI value of the forecast date,the prediction of the EVI value is completed.The prediction accuracy of the EVI-RBF model constructed in this paper is above 90%,and the trend of the prediction curve is consistent with the actual curve.(2)Research on crop yield forecast.In the research of completing the maize yield forecast,aiming at the problem of the failure of the traditional linear regression model,this paper proposes a method to use RBF neural network to complete the forecast of maize yield.The RBF neural network was used to establish the nonlinear relationship between the vegetation index time series and the yield during the key growth period of maize.The linear regression model,RBF model and BSSO(Bound Social Spider Optimization)-RBF model are used to verify the applicability of EVI,NDVI(Normalized Difference Vegetation Index),RVI(Ratio Vegetation Index)and SAVI(Soil-Adjusted Vegetation Index)in production forecasts.The results showed that the BSSO-RBF model achieved the best prediction accuracy and the highest correlation with production.Meanwhile,EVI presents the highest applicability in the production forecasting process.(3)Research on BSSO algorithm to optimize RBF neural network.When using the RBF model to predict maize yield in this paper,because the acquired remote sensing data is only 20 years old and the number of sample sets is small,the RBF model tends to fall into a local optimum during training,and it is difficult to guarantee the prediction accuracy.Therefore,this paper makes corresponding modifications based on the SSO(Social Spider Optimization)algorithm and proposes the BSSO algorithm.Further,BSSO algorithm is used to complete the optimization of RBF neural network.The research conducted comparative experiments on three common data sets to verify the optimization performance of BSSO algorithm and SSO algorithm,particle swarm optimization(PSO)and another social spider algorithm-SSA(Social Spider Algorithm).The experimental results show that the BSSO algorithm has more excellent performance in the number of iterations,prediction accuracy and stability in all data sets.This paper mainly completed the research of crop growth prediction and the research of using RBF model and BSSO-RBF model forecast yield when the linear regression model can’t forecast yield.It not only realizes the ability to monitor crop growth from both real-time and process perspectives to assist relevant departments in judging the pros and cons of growth,but also to complete high-precision forecasts of vegetation index and yield,thereby helping relevant departments to regulate related agricultural production activities in advance and formulate efficient food policies in advance.The research of subject has good application value.更多还原