Cotton Biomass Inversion Based On UAV Multi-spectrum

Cotton is a special commodity related to the national economy and people’s livelihood,and cotton is also an important commodity related to agriculture and textile industry.In recent years,the use of remote sensing technology to monitor the growth parameters of cotton has become a research hotspot in the field of agronomy.On this basis,this research experiment was conducted in the agronomy teaching and research practice base of Tarim University in Alar,Xinjiang.The canopy multi-spectral image data of cotton"Tahe-2"was obtained by UAV,the aboveground biomass of cotton was collected during the whole period,and the coordinate information of the collected samples was recorded.The correlation analysis between the cotton aboveground biomass and the band was carried out,and the vegetation index was selected.Finally,according to the selected vegetation index,the inversion model of cotton aboveground biomass based on partial least square method(PLSR),support vector machine(SVM),random forest regression(RFR)and extreme gradient lifting(XGBoost)was constructed respectively.The above ground biomass of cotton can be retrieved timely and accurately by multi-spectral inversion of unmanned aerial vehicle with cotton plant intact,which provides certain technical and theoretical support for cotton crop parameters and growth monitoring.The main work and achievements of the research are as follows:(1)Analysis of the original multi-spectral data showed that the spectral absorption valley was in the red band,while the reflection peak was in the green band.From the bud stage to the flower and boll stage of cotton,the reflectance of the spectral band increased significantly,because the biomass increased significantly with the influence of leaf chlorophyll content,leaf cell structure,water content and other factors,and the spectral reflectance of different bands showed a trend of gradual increase.From flower-boll stage to batting stage,the nutrients of cotton plants need to supply cotton peach,and the color of cotton leaves no longer changes significantly,nor does LAI continue to grow larger.Therefore,the reflectance of cotton plants at batting stage shows an obvious downward trend.(2)The correlation coefficient method was used to screen and invert the optimal multispectral bands of cotton aboveground biomass.The correlation coefficients of multispectral images of cotton in the four periods were strongly correlated with the correlation coefficients of cotton biomass in the four periods in the650 red band and 840 near infrared band,and the correlation coefficients of cotton biomass in the five bands were the highest.The correlation coefficients of the four periods of 650 red band are-0.66,-0.72,-0.66,-0.61,and 840 near infrared band are 0.67,0.64,0.60,and 0.44,respectively,and the correlation coefficients between 650 red band and 840 near infrared band are small,which conforms to the principle of band selection.Therefore,650 red band and 840 near infrared band were selected as vegetation index bands.Three planting indexes were selected according to the selected spectral bands:normalized vegetation index(NDVI),ratio vegetation index(RVI)and wide dynamic range vegetation index(WDRVI).(3)According to the selected three planting indices,they were used as input vectors for cotton above-ground biomass inversion models based on partial least square method(PLSR),support vector machine(SVM),random forest regression(RFR)and extreme gradient Lift(XGBoost).An inversion model of cotton aboveground biomass based on PLSR,SVM,RFR and XGBoost was constructed.The results showed that the best inversion model of seedling biomass was the RFR with NDVI as the input vector,the model accuracy coefficient R~2 was 0.845,and the root mean square error RMSE was 0.626.The accuracy of XGBoost inversion of bud biomass with RVI as the input vector was the best.The model accuracy determination coefficient R~2 was 0.894,and the root mean square error RMSE was 0.423.The best inversion model of flowering and boll biomass was XGBoost with WDRVI as input vector.The accuracy coefficient R~2 was0.858,and the root mean square error RMSE was 0.549.The best inversion model of biomass at the flocculating stage was XGBoost with WDRVI as the input vector.The accuracy determination coefficient R~2of the model was 0.894,and the root mean square error RMSE was 0.498.It can provide some theoretical and technical support for rapid and accurate acquisition of aboveground biomass of cotton so as to monitor cotton growth and field management.