Machine Learning-Based Hyperspectral Estimation Of Potato Yield

Potatoes are high-quality food crops and economic crops.Compared with other crops in the Yinshan area of Inner Mongolia,potatoes are more suitable for local meteorological conditions,and the sown area accounts for a large proportion.The tuber yield of potato is mainly affected by the balance of aboveground biomass(source)and belowground tuber dry matter(sink),and the estimation of aboveground biomass at the regional scale can provide a basis for monitoring and forecasting crop yield.Based on the hyperspectral data of potatoes in the area along the Yinshan Mountains,this study determined the relationship between above-ground biomass and tuber dry matter accumulation in potatoes.The biomass estimation model of potato tubers in Zishan area is based on the optimal spectral index and optimal growth period selected to initially establish the potato yield estimation model in this area.The linear relationship between aboveground biomass and tuber biomass of potato in the study area was well fitted,and tuber biomass had a good explanation for aboveground biomass.Compared with the published spectral index,the optimized spectral index significantly improved the modeling effect of potato tuber biomass.The optimal spectral indices screened out according to different growth stages are not the same.The spectral indices selected in the tuber formation stage are NDVI,ENDVI,BNI,NDDA and m RER;the optimal spectral indices NDVI,ENDVI,m ND705,BNI and m RER are selected in the tuber expansion stage;starch accumulation The optimized spectral indices selected for this period include RVI,ENDVI,m ND705,BNI and m RER,and the band combination positions are the red band(620-780 nm)and the near-infrared band(780-1 500 nm).Based on the optimized spectral index coupled with random forest model and partial least squares method,the Zhuozishan potato tuber biomass model was established.The independent variable was the optimized spectral index selected in three growth stages,and the dependent variable was the tuber biomass.The best tuber biomass estimation model was y=0.60x+714.72(R~2=0.79,RMSE=281.41kg/ha)in tuber expansion stage;y=0.91x+218.27(R~2=0.91,RMSE=257.27kg/ha)in starch accumulation stage.Among them,the tuber expansion period and starch accumulation period had better results,so the tuber expansion period and starch accumulation period were selected as the estimated growth period.After the optimal spectral index and the best estimated growth period were screened out,a yield estimation model of potato tuber expansion period and starch accumulation period was established based on this.The optimal yield estimation model was y=0.52x+1741.96(R~2=0.78,RMSE=199.73kg/ha)in tuber expansion stage;y=0.58x+1525.25(R~2=0.70,RMSE=258.12kg/ha)in starch accumulation stage.Based on the above results,the feasibility of using machine learning algorithm,namely random forest algorithm and partial least squares method to estimate potato yield based on hyperspectral index is proved,and it is shown that the use of spectral data can better estimate potato yield in tuber expansion stage and starch accumulation stage.Predicting potato growth.The results have guiding significance and important guarantee for improving the monitoring ability of potato growth and ensuring the stable increase of potato production.