Evaluatied Of Mulch Film Degradation And Residue,and Cotton Yield In Cotton Field Based On UAV Images

Mulch films were one of the key technologies for crop cultivation in arid areas of Northwest China,which had the functions of increasing temperature,preserving soil moisture and increasing yield.With the increasing scale of mulch film use years by years,the residue pollution of mulch film is becoming increasingly serious.Therefore,accurate and efficient evaluation of mulch film residue is the great practical significance to guide the accurate recovery of mulch film and to the promotion of high-yield and environmentally friendly mulch films.In this paper,seven treatments of the high strength all recovery mulch film（GB）,polyethylene mulch film（PE）,no-cover mulch film（JM）and four kinds of degradable mulch films（TZ,LS,BF and B）were used to cover cotton fields.The remote sensing images of cotton fields were obtained by UAV.Explored the identify effects of the Maximum Likelihood Classification（ML）,the Minimum Distance Classification（MD）and the Sectral Angle Mapper Classification（SAM）on different types of mulch film images.Based on six machine learning model algorithms,constructed the estimation models of the mulch film residue、the mulch film degradation rate,and the cotton above-ground biomass（AGB）and yield.Comprehensive the degradation and residue of mulch film,and its influence on cotton yield,finally realized the evaluation of mulch film degradation and residue,and cotton yield based on UAV image.The mian conclusions can be summarized as follows:（1）In this paper,ML,MD and SAM were used to feature identify the high strength all recovery and the degradable mulch films.The correlation coefficients between the identification results and the measured data were 0.9,the determination coefficient（R²）of the residue estimation model was about 0.9,and that of the degradation rate estimation model was about 0.8.ML had the best identify effect from the identify results,correlation analysis and estimation model accuracy.（2）The Bayesian Ridge Regression（BRR）and Partial Least Squares Regression（PLSR）models had high accuracy in the estimation of residue and degradation rate.The R² of the estimation model was 0.8～0.9,and the verification was 0.7～0.8.The accuracy of different mulch film types of optimal models was different,but the classifier identifity results tended to be linear with the measured data.（3）Estimation of GB and PE Residues,and LS,BF,B and TZ degradation rates based on the optimal model of mulch film type,the errors were less than 0.2 with the measured results.（4）The Support Vector Regression（SVR）algorithm had the best effect on constructing the estimation model of cotton Above Ground Biomass（AGB）at flowering stage,with the estimation model（R²=0.73,RMSE=0.21,r RMSE=9.98%）and verification（R²=0.80,RMSE=0.21,r RMSE=10.05%）.The PLSR algorithm had the highest response to the cotton yield estimation model,with the estimation model（R²=0.77,RMSE=0.05,r RMSE=8.73%）and verification（R²=0.67,RMSE=0.06,r RMSE=10.54%）.（5）Comprehensive the analysis of mulch film degradation rate and yield,the yield of GB treatment was 3.66%higher than that of PE,and the residue was reduced by 18.46%.The estimated results respectively were 0.567 kg·m^-2 and 1.3 g·m^-2,which had the effect of increasing yield and environmental protection,and more in line with market demand.Comprehensive the analysis of mulch film degradation rate and yield,the yield of TZ treatment was the first,and the yield of LS treatment was reduced by 3.65%,which was 0.502 kg·m^-2,but the degradation rate was increased by 12%,which was 79%.LS had good application potential.