Rice Quality Estimation Based On Multi-platform Hyperspectral Remote Sensing Data

As rice is one of the main cereal crops in China,its safe and high-quality production activities play an extremely important role in the stability and development of the country.In the wake of economy’s development,the public put forward higher requirements for the quality of rice.The amylose content,protein content and total starch content of rice grains are all important indicators for evaluating rice quality,which directly or indirectly affect the taste and nutritional quality of rice.In this paper,the ASD spectrometer is used to collect indoor spectra of of rice grain,brown rice and rice flour,as well as the spectral reflectance of rice canopy at key growth stages.The DJI M600 Pro hexarotor UAV equipped with the Rikola imaging hyperspectrometer is used to obtain the canopy of rice at key growth periods.Based on this,three quality monitoring models of the scales of indoor sampeles,canopy,and drone have been established to predict and verify the quality index parameters such as amylose content,protein content and total starch content of rice grains,and try to further optimize the UAV multi-growth period quality monitoring model by coupling vegetation index and UAV image texture parameters.The following are the specific findings of the research:(1)At the indoor sample scale,we selects indoor ASD spectra of rice samples in the form of rice grain,brown rice and rice flour,and establishes an estimation model of rice grain quality indicators such as amylose content,protein content and total starch content through correlation analysis and multiple stepwise regression.It is found that the rice flour spectrum is more suitable for the prediction of rice quality indicators based on the indoor spectrum than that of the brown rice and rice grain.The increase in the processing degree of rice samples is beneficial to the improvement of the quality prediction ability.The best indoor spectral prediction models for the three quality indicators are all based on rice flour spectrum.The modeling determination coefficients of the prediction models for amylose content,protein content,and total starch content reach 0.6891,0.6231 and 0.8423,respectively.(2)At the canopy scale,we collect ASD spectra of rice canopy at booting period,heading period,milkling period and maturity period,and establish prediction models of amylose content,protein content and total starch content in rice grains based on the quantitative combination of different growth stages through correlation analysis and multiple stepwise regression.A comprehensive comparison of the modeling and verification effects of different growth period combination models shows that the increase in the number of growth periods can significantly and comprehensively improve the accuracy,reliability and stability of the quality monitoring model.The best models for monitoring experiment of amylose content,protein content and total starch content are all the combined model of booting,heading,filling,and maturity stages.The modeling R2of the best model for amylose content,protein content,and total starch content is 0.8945,0.7661,and 0.8276,respectively.(3)At the drone scale,we obtained the UAV imaging hyperspectral images of the rice booting stage,heading stage,filling stage and maturity stage in the study area.A multi-growth-stage rice quality monitoring model is constructed based on UAV hyperspectral image through correlation analysis and multiple stepwise regression methods.The study found that although its effect is not as good as the introduction of multiple growth period information at the canopy scale,and the number of growth periods that can be introduced is relatively limited,the comprehensive utilization of multiple growth period information can still improve the effect and stability of UAV rice quality prediction to a certain extent.The monitoring model of rice grain amylose content based on UAV spectrum is a combined model of heading,filling,and maturity stages,and its modeling R~2=0.8772.The monitoring models of rice grain protein content and total starch content based on UAV spectroscopy are both combined models of two growth stages at booting stage and heading stage,and their modeling R~2 is0.6835 and 0.8456,respectively.(4)UAV hyperspectral images have the characteristics of spectral and texture fusion,and try to introduce spatial texture information to improve thespectral prediction model.The gray-level co-occurrence matrix method is used to extract the spatial texture information of the UAV images of each growth period,and the UAV rice quality prediction model that couples the vegetation index and texture information of the multiple growth periods is constructed through correlation analysis and stepwise regression.The study found that the introduction of spatial texture information can fully optimize the model accuracy,reliability and stability of the starch quality indicator monitoring model.The improvement of the amylose model is particularly significant,but the effect of protein indicators is not good,which will lead to a decrease in the stability of its model.Therefore,the unmanned aerial vehicle rice quality prediction technology that couples vegetation index and spatial texture information is only applicable to starch indicators.