Study On Rice Yield Estimation Model Based On Multispectral Image

Rice is one of the three major food crops in China,and also one of the main consumption rations for Chinese residents.Rice yield is related to agricultural economy.It is of great practical significance to estimate rice yield scientifically and accurately for agricultural economic development.At present,with the development of precision a griculture,people put forward higher requirements on the estimation accuracy and cost of rice production.An important way to improve the accuracy of rice yield estimation is to improve the method of rice yield estimation effectively.In this study,the canopy multispectral images of rice in the experimental field were obtained by using a multi-spectral camera mounted on a small UAV platform.Based on this multi-spectral image,the corresponding research and attempt on rice yield estimation methods were carried out.Taking rice in Qing'an County of Heilongjiang Province as the experimental object,based on the multi-spectral images of rice canopy at jointing stage,the characteristics of rice canopy vegetation index extracted from the images were taken as t he research object,and the characteristics of rice growth parameters collected synchronously as the control object,the relationship between the two characteristics and rice yield was studied.Relevance test was used to screen the traits,and the traits with low correlation with yield traits were eliminated.Six selected characteristics were used as yield estimation factors,and local weighted linear regression and quantile regression were used to model rice yield estimation.Mean square error(RMSE)and mean absolute percentage error(MAPE)are used as test indexes to analyze the accuracy of different yield estimation models constructed by the two algorithms.The main research contents are as follows:(1)According to the characteristics of the follow-up Rice Yield Estimation Modeling Algorithm in this study,the linear correlation between the characteristics of rice growth parameters,vegetation index and rice yield was tested by normal test and Pearson correlation coefficient test.Normality test results show that the distribution of all characteristic data is normal or close to normal distribution,which meets the requirements of Pearson correlation coefficient test method.The results of correlation test showed that the correlation coefficients with yield characteristics were below 0.5,which were divided into plant height and leaf temperature in growth parameters and RVI in vegetation index characteristics.Six effective yield estimation factors were screened out,which could be used in subsequent rice y ield modeling.It includes two characteristics of rice growth parameters(SPAD,LNC)and four characteristics of rice canopy vegetation index(NDVI,DVI,SAVI,OSAVI).(2)Using the selected yield estimating factors,the rice yield estimating models were constructed by using local weighted linear regression and quantile regression respectively.The results showed that in the local weighted linear yield estimation model of rice constructed in this study,the effect of using SPAD to construct the rice yield e stimation model was the best.The determination coefficient R2 was 0.72,the average absolute percentage error MAPE was 3.75,and the root mean square error RMSE was 18.8 kg/mu.The best vegetation index characteristic was OSAVI,and its determination coefficient R2 was 0.69,and the average absolute percentage error was 3.75.MAPE was 3.53,RMSE was 19.5 kg/mu.(3)Among the quantile regression yield estimation models constructed in this study,NDVI based on vegetation index has the best effect,with the d etermination coefficient R2 of 0.675,the average absolute percentage error MAPE of 4.203 and the root mean square error RMSE of 44.71 kg/mu.Under the_=0.7 locus,the quantile regression model y=-46.72x+665.96 constructed by NDVI and rice yield characteristics is the best quantile regression yield estimate on model for rice.