Study Of Discrimination And Analysis Of Hyperspectral Characteristics Of Typical Vegetation In Rare Earth Reclamation Mining Area

In recent years,hyperspectral remote sensing technology has been widely used in the fine recognition of vegetation types in agriculture,forestry and grassland because of its large number of bands and nanometer spectral resolution.It solves the problem that traditional remote sensing can only identify large types of land objects.However,different vegetation types have different constraints due to different regions,so they have special spectral characteristics for vegetation.The selection of parameters and analysis methods are also different,and the vegetation in mining area is seldom studied because of the regional particularity.Large-scale mining of ion-type rare earth mining areas in southern China has resulted in deterioration of ecological environment,such as soil erosion and vegetation destruction,etc.Its main control methods rely on artificial reclamation technology,but reclaimed vegetation is difficult to absorb nutrients from damaged soil,resulting in generally poor growth and unsatisfactory reclamation effect.At the same time,mixed planting method is often used to reclaim vegetation in mining areas.It is difficult to retrieve vegetation physical and chemical parameters from hyperspectral remote sensing.In this paper,hyperspectral data of four typical reclamation vegetations(Photinia fraseri,Pinus massoniana,Vernicia fordii and Salix fragilis)in the Aubeitang Rare Earth Mine of Dingnan County,Ganzhou were measured,their spectral curve characteristics were studied,and the recognition method of reclamation vegetation was constructed,which could provide experimental basis for vegetation inversion,monitor the growth trend of typical reclamation vegetation in the ion rare earth mining area of southern China and reflect the reclamation effect of the mining area.It provides theoretical basis and technical support.Firstly,the measured hyperspectral reflectance data of four typical reclamation vegetations are pretreated,and the spectral differences between them are amplified by using first derivative(FDR),reciprocal logarithm(Log(1/R))and de-envelope(CR)spectral transformation techniques.Then,the dimensionality of vegetation spectral data is reduced based on the method of Mahalanobis distance and mean confidence interval band.Finally,Fischer's method is used to reduce the dimensionality of vegetation spectral data.Four kinds of reclaimed vegetation are identified by Fisher and Bayes methods,and the classification accuracy is analyzed and compared.The main conclusions are as follows:(1)The spectral curves of four kinds of vegetation have basic peak-valley characteristics,which conform to the trend of green vegetation spectral curves with good growth.There are two valleys and one peak and red edge in the visible band.Because of the influence of environmental factors in the mining area,there are different degrees of "blue shift" in the four vegetations,with Pinus massoniana and Bamboo Willow being the most obvious.Four kinds of reclaimed vegetation have different trilateral parameters in derivative spectrum,"two peaks and one valley" in Log(1/R)and absorption depth,K and S in CR,which indicates that using these spectral transformation techniques to distinguish reclaimed vegetation types is effective.(2)The spectral characteristic bands of reclaimed vegetation selected by the mean confidence interval band method based on the optimal dimension reduction method in this paper are 536-578 nm of the original spectrum,530-541 and 1032-1065 nm of the reciprocal logarithmic spectrum,396-492,536-558,639-687 and 1110-1143 nm of the de-envelope spectrum.(3)Based on Bayes and Fisher discriminant methods,the vegetation is recognized effectively by the spectral transformation of each reclaimed vegetation.After spectral transformation,the classification accuracy of the four vegetation types is improved compared with the original spectrum.The overall classification accuracy of the two methods in the de-envelope spectrum is 0.806 and 0.813,respectively,which are better than other transform spectra.As far as the original spectrum is concerned,Fisher discriminant method has a higher overall classification accuracy,and Bayes method is better than Fisher method for the transformed spectrum(reciprocal logarithmic spectrum and de-envelope spectrum).