Research On Monitoring Model Of Soil Salinization Based On Multispectral Remote Sensing Of UAV

Due to the influence of human activities and natural conditions,soil salinization in Hetao Irrigated Area is becoming increasingly serious,which brings great challenges to the sustainable utilization of soil resources.UAV(Unmanned aerial vehicle)remote sensing can quickly obtain a wealth of features spectral information,centimeter-level spectral resolution accurately reflects the features of the features spectrum,which is an important means for accurate and real-time monitoring of soil salinity.In this paper,the Shahaoqu Irrigation Area in Hetao Irrigation Area was taken as the study area.This reaserch obtained the soil salinity information of the cultivated land with different degrees of salinity,and obtained the UAV multi-spectral remote sensing images of the objects in the study area during the bare soil period and the vegetation cover period.The characteristics of the spectral curves of the ground features in the two periods were analyzed,and through the obtained spectral band reflectance,a variety of spectral indexes are constructed,and the correlation between the characteristics of the spectral curve of the ground features and the multi-spectral data(band reflectivity and spectral index)with the soil salt content in the two periods are analyzed,subsequently univariate linear regression models based on sensitive bands were established.During the vegetation cover period,soil salt content prediction models were established using different model input variable groups coupled with multiple regression methods.in the bare soil period,mathematical models were established based on variable selection methods combined with machine learning algorithms to quantitatively predict soil salinity.After comprehensive evaluation by multiple evaluation indicators,the prediction model of soil salinity was finally determined(1)Study on spectral characteristics of ground objects and correlation of salt spectrum based on remote sensing data The difference between the multispectral curves of bare soil and vegetation canopy is obvious.The reflectance of vegetation canopy is low and the reflectance of bare soil is high.Both types of curves show a rising trend with wavelength increasing and with the degree of saline soil increases,the reflectance also increases.The four multispectral bands(490nm,550nm,680nm,800nm)in the bare soil period showed a significant correlation with the soil salinity content,while the three bands(680nm,800nm,900nm)in the vegetation canopy during the vegetation cover period showed a significant correlation with the soil salinity content relationship.The prediction accuracy of the unitary linear regression model in both periods is poor.Relatively speaking,the accuracy of the salt linear regression model in the bare soil period is higher than that in the vegetation cover period.(2)Predicting soil salt content using different variable groups coupled with multiple regression methods in vegetation cover period.By analyzing R~2(Coefficient of determination)and RMSE(Root mean square error)of three variable groups,the spectral index group has achieved the best inversion results in the four regression model methods.The inversion effect of the sensitive band group and the full variable group in different regression algorithms is different.Among the four regression methods,the accuracy of the inversion of the three machine learning regression algorithms is significantly higher than that of the muti linear regression model(MLR)model,and the"overfitting"phenomenon appears in the sensitive band group and the full variable group in the MLR model.The Random Forest(RF)algorithm performs best among the three machine learning algorithms,while the support vector regression(SVR)algorithm and back propagation neural network(BPNN)algorithm perform differently in models based on different variable groups.The RF model based on the spectral index group achieved the best prediction results among the twelve models.R_c~2 and R_v~2 reached 0.72 and 0.67,respectively,and the RMSE_v was only0.112%.(3)Predicting soil salt content using variable selection method combine with machine learning algorithms in bare soil period.The results showed that the estimation accuracy of the models had been improved markedly using three variable selection methods,and VIP outperformed GRA and GRA outperformed SPA.However,the model accuracy with the three machine learning algorithms turned out to be significantly different:RF>SVR>BPNN.All the twelve SSC estimation models could be used to quantitatively estimate SSC(RPD>1.4)while the VIP-RF model achieved the highest accuracy(R_c~2=0.835,R_P~2=0.812,RPD=2.299).