Research On Soil Water Content Inversion Model Based On Hyper Spectral Data

Soil water content is one of the important factors affecting crop yield and soil and water conservation.Mastering its changes can provide reliable information for research and protection of soil,which is also crucial for precision agriculture.Compared with the shortcomings of traditional chemical analysis methods,such as huge labor cost and high time cost,hyperspectral technology has the advantage of monitoring soil moisture content in a large area,quickly and accurately.Therefore,the introduction of spectral data into the soil water content inversion model can more accurately and quickly monitor the dynamic changes of soil water content in the region,and provide new ideas and methods for the inversion of soil water content in the region.In this study,the soil of the teaching and research base of Jilin Agricultural University was taken as the research area.The soil samples were collected by five-point sampling method and ground.The spectral data were obtained by ASD spectrometer,and the water content data were measured by drying method.Based on this,the soil water content inversion model was studied in the study area.The specific research work is as follows:(1)In this study,the soil of the teaching and research base of Jilin Agricultural University was taken as the research area.The soil samples were collected by five-point sampling method and ground.The spectral data were obtained by ASD spectrometer,and the water content data were measured by drying method.Based on this,the soil water content inversion model was studied in the study area.(2)The sample point data based on the characteristic band is eliminated by principal component analysis combined with Mahala Nobis distance,and then the modeling set and test set are divided by K-S algorithm.Finally,the BP neural network model and support vector regression machine model are constructed by using the modeling set data,and the optimization parameters of the model are adjusted respectively.The results of the test set are evaluated by~2and RMSE evaluation indexes.The BP neural network model(~2=0.9239,RMSE=0.05)is superior to the support vector regression model(~2=0.42842,RMSE=0.061),which indicates that the BP model can be used as a high-precision model for soil water content inversion.(3)A BP neural network model based on sparrow search optimization strategy is proposed.This paper improves by using the method of calculating fitness and sorting.The improved model improves the recognition accuracy by 2.71%(~2=0.9510,RMSE=0.038),and further shortens the training time,improves the learning efficiency and accelerates the convergence of training.Through evaluation,it is considered that the model can be used as a model with high precision inversion ability in soil water content inversion experiment.