Lai Inversion Of Winter Wheat Based On UAV Hyperspectral Remote Sensing

As one of the main crops in China,winter wheat occupies an important proportion in food consumption of our country.Leaf area index(LAI)is one of the most important biochemical parameters of crop growth and dynamic monitoring index to the reflection of size and growth of crop population.In recent years,with the rapid development of remote sensing technology,sensors tend to the direction of higher spatial resolution and higher spectral resolution.Although satellite remote sensing can obtain a wide range of remote sensing data based on the ground,because of the climatic conditions and revisiting period,data’s effectiveness is not rich.Because of the flexibility and less affection by the atmospheric effects and rich information covering of the ground,hyperspectral remote sensing based on UAV has become an important means of data acquisition of the crop growth monitoring.Although empirical model for LAI inversion is simple,it makes a big loss of the effective information of hyperspectral remote sensing data.Therefore,this paper proposes a method of LAI inversion based on dimension reduction of UAV hyperspectral remote sensing data.Taking National Precision Agriculture Experiment Station in Beijing Changping District as research area,acquisition main growth period of winter wheat including Jointing stage,Flag picking stage and flowering stage of hyperspectral remote sensing data by UAV and synchronous ASD and LAI data based on ground,this paper finished the following research work:(1)Analysis of correlation between adjacent bands and pixels of the main growth period;(2)Using hyperspectral dimensionality reduction toolbox including PCA,LE,LPP,LDA and Isomap algorithm based on MATLAB Platform to finish dimensionality reduction of UAV hyperspectral remote sensing data of main growth period;(3)Building LAI inversion regression model between dimensionality reduced hyperspectral data and synchronous LAI data by support vector machine(SVM)optimizated by cross validation;(4)comparative analysis of empirical model and inversion model based on LAI dimensionality reduction algorithm.The results show that:①There is a strong nonlinear correlation between adjacent spectral bands in hyperspectral remote sensing image,and the effective information can be extracted by dimensionality reduction.Different dimensionality reduction algorithm of hyperspectral image has different information extraction ability.Linear dimensionality reduction algorithm has slow convergence speed and weak noise elimination ability.The dimensionality reduction algorithms based on manifold learning show fast convergence and noise elimination ability.Global dimensionality reduction algorithm is better than the local dimensionality reduction algorithm in information maintaining.②To get better inversion results,the parameters of support vector machine is optimized by cross validation using the feature dimension with high correlation with LAI.Isomap algorithm has the highest accuracy of all three phase LAI inversion,R2 of three phase are 0.7813,0.8171,0.8017,and the RMSE are 0.4745,0.5291,4927,The highest accuracy of linear dimension algorithm of three phase inversion is PCA,and R2 are 0.6866,0.7482,0.7865,RMSE are 0.5271,0.6721,0.5089.The overall accuracy of Isomap algorithm is higher than that of the PCA algorithm.③LAI inversion method based on empirical model is simple to operate,easy to understand,but with the loss of effective information,the overall accuracy is not very high.Because of effective eliminate noise,data reduction,information extraction ability,the LAI inversion precision can be improve by global nonlinear dimensionality reduction.hyperspectral image can be treat as the data embedded in a lower manifold surface.