Font Size: a A A

Research On The Estimation Algorithm Of Rice Leaf Area Index Based On Sentinel-2 Remote Sensing Data

Posted on:2022-12-01Degree:MasterType:Thesis
Country:ChinaCandidate:C H QiFull Text:PDF
GTID:2512306614455344Subject:Automation Technology
Abstract/Summary:PDF Full Text Request
It is vital to obtain rice leaf area index quickly and accurately for assessing rice growth and enhancing the efficiency of field management.At present,most studies only use two or three bands in the spectral band,or construct vegetation index or use machine learning to fit LAI.First of all,it is difficult to fully represent the change of biophysical parameters with fewer bands.Second,vegetation index,simple linear fitting and shallow machine learning cannot fully explore the change rule implied by the information,and the inversion accuracy is poor.To address these issues,this paper proposed a method to extract the characteristic bands by two-dimensional correlation spectroscopy,and then use the gated cyclic unit-support vector regression inversion model(GRU-SVR)to invert rice leaf area index.In order to extract sensitive bands that can represent rice lai as comprehensively as possible,this paper used PROSAIL model to generate simulated rice canopy spectral data with LAI as external disturbance,and then carried out two-dimensional correlation spectral analysis.Nine bands of 465 nm,530 nm,565 nm,580 nm,705 nm,765 nm,790nm,895 nm and 945nm were extracted as sensitive bands of lai.Compared with stepwise multiple linear regression,the characteristic bands screened by two-dimensional correlation spectroscopy were more specific and could better represent the leaf area index of rice.As a widely used algorithm,support vector machine method has outstanding strengths in processing nonlinear and high-dimensional data,but it cannot make full use of the characteristics of sequential data and involves many parameters,resulting in low accuracy and long running time of the model finally established.Recurrent neural network has outstanding performance for sequential data,but it needs a lot of label sample to support,and it is hard to obtain in general.Therefore,in order to take full advantage of the two algorithms,a rice leaf area index inversion model based on GRU-SVR was proposed in this paper.By establishing a rice leaf area index inversion model based on SVR,it was proved that the two-dimensional correlation spectroscopy has outstanding advantages in the field of feature band extraction;SVR inversion model based on radial basis kernel function has the best performance;adding 70%historical measured data can improve the universality of the model.Based on the above conclusions,the proposed model has higher accuracy(R~2=0.93354,RMSE=0.52283)than other models,and is more suitable for the actual rice leaf area index inversion task.The sentinel-2 satellite remote sensing images were input into the trained model to generate the graded distribution map of rice leaf area index in the study area.It makes it possible to master the growth situation of rice from a macro perspective and provides a guiding direction and a broader perspective for subsequent field management.
Keywords/Search Tags:Remote sensing, Rrice leaf area index, Feature band extraction, Machine learning
PDF Full Text Request
Related items