Rice Yield Estimation Based On Time-series Visible Unmanned Aerial Vehicle Remote Sensing Images

Rice is one of the three major food crops in the world,the timely and accurate prediction of rice yield is of great significance to guide national agricultural production and ensure national food security.In recent years,the rise of unmanned aerial vehicle(UAV)applications has provided new opportunities for agricultural remote sensing monitoring.How to use the time-series visible UAV remote sensing images to quickly and accurately extract rice planting area and achieve rice yield estimation is a necessary condition for precision agriculture.Taking the Jianghan Plain,a typical rice planting area in central China,as the research area,the paper aims at rice yield estimation.Combined with the characteristics of large number of plots,uneven area,complex terrain and many crop types,we focus on the application of parcel boundary extraction,paddy rice coverage extraction and rice yield estimation.The main research contents and contributions of this paper are as follow:(1)Combined with the characteristics of rice growing areas in the study area,the process of automatic extraction of parcel boundary is proposed with the multiscale combinatorial grouping segmentation algorithm,and we study the effect of spatial resolution and segmentation scale on the accuracy of parcel boundary extraction.Based on high resolution agricultural UAV images,we use the optimal spatial resolution and optimal segmentation scale to achieve automatic extraction of agricultural land parcel boundaries.Experiment results show that the most suitable ground sampling distance for extracting land parcel boundary is about 30 cm and the optimal segmentation scale is [0.2,0.4].The accuracy of land parcel boundary extraction can be more than 90%.Moreover,parcel boundary can provide the necessary spatial basis for paddy rice coverage extraction and yield estimation.(2)Considering the practical application requirements of extracting planting area in the early stage of rice growth,a rule-based hierarchical classification framework is constructed to realize rice planting area extraction.The classification threshold is determined by selecting the optimal phase and the visible vegetation index with the ROC curve,the decision rules for hierarchical classification are constructed.And the parcel is used as the spatial unit of classification based on the extraction of boundary.The experiment results show that the hierarchical classification framework combined with threshold can quickly and accurately extract paddy rice coverage,and the overall extraction accuracy is over 95%,which can provide crop spatial distribution information for rice yield estimation research.(3)Based on the extraction of parcel boundary and paddy rice coverage,the visible UAV remote sensing images are used to estimate the yield.Aiming at the time series images of multiple growth stages,we use the method of multiple linear regression and random forest regression to establish the rice yield estimation model of single growth period and multiple growth period respectively.The effects of estimated period and visible vegetation index on rice yield estimation accuracy are studied.The model is analyzed and evaluated by the measured yield data,and the rice yield estimation is realized in the experimental area of Zhijiang City,Hubei Province.The related work of this paper is based on the time-series visible UAV remote sensing images,and we research and process the relevant data and technical processes,including plot boundary extraction,rice planting area extraction and rice yield estimation,which can provide the research basis and theoretical model for rice yield estimation in the future.