Research On Inversion Of Leaf Area Index Based On Absorption Peak Characteristics

Hyperspectral material inversion is an important research direction for applying hyperspectral remote sensing technology to practical production and life.High-precision inversion algorithm is helpful for researchers to obtain more accurate ground material information.Leaf area index(LAI)is a common parameter in the research of vegetation growth in a given area.Traditional methods for inversion of leaf area index based on hyperspectral image mainly use the reflectance of a band or an index generated by the combination of several bands,whose inversion accuracy is generally limited.In order to improve this problem,considering the characteristics of the absorption peak of the spectrum can invert the structural characteristics,this thesis focuses on using the multivariate absorption features to invert the leaf area index of the vegetation.The main content of this thesis is as follows:(1)This thesis introduces the basic knowledge of hyperspectral inversion and the current research status,and points out the common methods of inversion of leaf area index and some advantages of using inversion features.(2)We propose to use the absorption features to invert the leaf area index.First,the absorption peak is extracted after the pretreatment of original spectrum.The empirical linearity,exponential,exponential,logarithmic,BP(back-propagation)neural network and support vector machine models are calculated for different absorption features and their combinations with leaf area index,benchmarked with the others based on the reflectance or vegetation index.Lastly,the test samples are used to examine the best model among them.The experimental results show that some of the features of the absorption peak outperform the benchmarked methods of spectral reflectance or spectral index in the same model in accuracy and error.(3)A multi-scale method is proposed to obtain the absorption features.The empirical linearity,exponential,power exponent and logarithmic models are established between the absorption peak characteristics and leaf area index at different scales.At the same time,the effect of the support vector machine regression model with different absorption features inversion of leaf area index is compared.The multi-scale methods are Fingerprint method and MMMT(maximum modulus wavelet transform)method.The experimental results show that theabsorption features obtained at different resolutions are different and choosing the best scale can improve the accuracy of the inversion model.The accuracy of the inversion model obtained by MMMT method is better than Fingerprint method.The absorption features in the visible range performers are best to the leaf area index retrieval.