Monitoring Of Growth Indexes Of Medium Indica Rice By Different Spectral Equipment Based On Random Forest Optimization

In this study,ground hyperspectral ASD equipment and unmanned aerial vehicle multispectral equipment were used as monitoring platforms.Five nitrogen fertilizer gradients of two varieties of medium indica rice（HLY 898 and YLY 900）were set:N0(0 kg·ha^-1),N1(75 kg·ha^-1),N2(150 kg·ha^-1),N3(225 kg·ha^-1)and N4(300 kg·ha^-1).The spectral information of each variety was systematically processed,and the relationship with the main agronomic parameters（Biomass,SPAD,and LAI）of medium indica rice was analyzed.The main findings of this study:（1）On the ASD platform,during the growth and development of rice,the vegetation index showed obvious saturation,which led to the decline of estimation precision and accuracy of agricultural parameters.In this experiment,the band depth analysis was performed on the original spectrum.The canopy spectrum in the range of550-750 nm was processed based on the band depth information（BD,BDR and NBDI）.The biomass,SPAD and LAI model were established.The monitoring accuracy（R²=0.68-0.83）,SPAD（R²=0.53-0.58）,and LAI（R²=0.59-0.67）had great correlation.（2）On the multispectral platform mounted on the unmanned aerial vehicle,the correlation analysis was performed on the agronomic parameters of each period and the whole growth period to find the optimal vegetation index on the unmanned aerial vehicle.The results showed that in biomass monitoring,the correlation was above0.58 in each period,and the correlation was 0.82 in multiple growth periods.On the monitoring of SPAD,the correlation was above 0.53 in each period,and the correlation was 0.57 in multiple growth periods.In LAI monitoring,the correlation was above 0.59 in each period,and the correlation was 0.73 in multiple growth period,showing a good correlation.The correlation between that vegetation index construct based on the multi-spectrum carry by the unmanned aerial vehicle and the agronomic parameters in the multi-growth period（R²=0.57-0.82）is significantly higher than that in each growth period（R²=0.53-0.59）.In the correlation analysis between the multi-growth period and the agronomic parameter,the correlations of the vegetation indexes GVI,R,G and B are better.（3）In order to further improve the accuracy of monitoring agricultural parameterswith two kinds of equipment,the agronomic indicators of medium indica rice in multiple growth stages were treated by mathematical methods(stepwise regression（SMLR）,small squares regression（PLS）,and random forest（RF）.On ASD,the prediction accuracy of biomass was improved to 0.89,that of SPAD was improved to0.89,and that of LAI was improved to 0.78 by the combination of random forest and band depth analysis.After the combination of random forest and UAV vegetation index,it was found that the prediction accuracy of biomass was improved to 0.92,the prediction accuracy of SPAD was improved to 0.76,and the prediction accuracy of LAI was improved to 0.87.In summary,the mathematical algorithm combined with ASD band depth analysis and UAV vegetation index modeling respectively showed that among the three mathematical models,random forest（RF）had the best effect on monitoring agricultural parameters,and the random forest could effectively improve the accuracy of monitoring agricultural parameters by remote sensing equipment.