A Method Of Determining Red Edge Position Of Rapeseed Spectrum Based On Multifractal Theory

With the continuous improvement and development of hyperspectral remote sensing technology,agricultural hyperspectral remote sensing technology is also more and more practical for crops or plants.Among them,plant parameters are closely related to the variation of plant hyperspectral bands.It is an important theoretical basis for plant research.In this paper,to study the rapeseed's canopy spectrum,a method of extracting red edge position of the spectrum was proposed by using local detrended fluctuation analysis(L-DFA).Also,the L-DFA method was compared with other six traditional methods,namely,maximum first derivative spectrum,inverted Gaussian fitting technique,linear four-point interpolation technique,Lagrange interpolation polynomial,polynomial fitting technique,and linear extrapolation technique.Take the local Hurst exponent of red edge position of rapeseed's canopy at pod period calculated by L-DFA and spectral values of red edge position obtained by other six methods as the features,a qualitative identification model was constructed by using two classifiers of support vector machine kernel method and random forest,to distinguish two plant modes of direct seeding vs.transplant,and two planting densities of 10 thousands per mu vs.20 thousands per mu.The results show that L-DFA is superior to other six methods because the former has best identification accuracy and stability.