Study On Vegetation Feature Extraction And Chlorophyll Content Retrieval In A Rare Earth Mining Area Based On Measured Ground Spectra

Ion-adsorbed rare earth is a rare and strategic resource in China,but long-term and uncontrolled mining in mining areas has caused serious damage to the surrounding ecology.Under artificial reclamation technology,the reclaimed land in mining areas still faces multiple ecological stresses,and the vegetation growth status is extremely poor.Therefore,it is necessary to monitor the growth status of reclaimed vegetation and accelerate the process of ecological restoration and governance.Through the obtained spectral data onto vegetation,fractal dimension calculation,wavelet transform analysis technology and Fourier transform processing in signal processing will be used to strengthen the detailed information of the spectrum,and the spectral characteristics of reclaimed vegetation will be analyzed.Compare with the first derivative and other commonly used transform processing,the contrast difference between reclaimed vegetation and normal vegetation will be more enhanced.Finally,this paper takes the sensitive bands of reclaimed vegetation in the original spectrum after pretreatment of reclaimed vegetation in mining area,the sensitive bands of vegetation in the first derivative spectrum(scales of d4,d5 and d6),the sensitive bands of vegetation in the original spectral wavelet transform spectrum and the spectral reflectance of the sensitive bands of vegetation in the first derivative wavelet transform spectrum(scales of d4,d5 and d6)as independent variables.The chlorophyll content of the corresponding reclaimed vegetation measured in the mining area was the dependent variable.The chlorophyll content inversion models were constructed by multiple stepwise regression method and back propagation neural network method,and the accuracy evaluation and effect comparison of the models were carried out according to~2 and.Finally,the following conclusions are reached:(1)After the first derivative transformation,the spectral curves of the reclaimed Photinia serrulate,the reclaimed Tung oil tree and the reclaimed Camellia on the red edge showed a large-scale blue shift.The blue shift of Camellia and Photinia serrulate reached a maximum of 18nm.It shows that the reclaimed vegetation in the mining area is affected by different degrees of environmental stress and other external factors,among which the reclaimed Photinia serrulate,reclaimed Tung oil tree and reclaimed Camellia have received greater environmental stress.At the same time,according to the calculation results of fractal dimension of vegetation spectral curve in the mining area,the fractal dimension of the same reclaimed vegetation is higher than that of the normal vegetation.(2)The vegetation leaf spectrum is amplified by discrete wavelet transform at multi-scale.The best detail coefficient of the original spectral discrete wavelet transformed is d5,and the best detail coefficient of the first derivative spectral discrete wavelet transform is d6.The spectrum is localized on the space-frequency map through short-time Fourier transform.The space-frequency characteristics of the original spectrum appear at the first trough of the red edge and mid-infrared,while the first-order derivative short-time Fourier transformed to amplify and increases the space-frequency characteristics of the spectrum curve at a smaller scale.(3)Based on the original spectrum,the first derivative spectrum,the original spectrum wavelet transformed spectrum and the first derivative wavelet transformed spectral reflectance combined with the chlorophyll content,the sensitive bands were extracted.The experimental results show that after the spectral transformation,its sensitive band is still concentrated on the visible light near infrared position.However,the original spectrum wavelet transformed spectrum and the first derivative wavelet transform spectrum are better than the original spectrum and the first derivative spectrum in both the number of sensitive bands and the correlation strength.(4)Through~2 and,the inversion model of chlorophyll content constructed by various methods is comprehensively evaluated.After model training,the overall inversion accuracy of BP neural network method is higher than that of multiple stepwise regression method,among which the reclamation of pine trees is the most obvious.The model of reclaimed pine trees was built by multiple stepwise regression method,with the highest~2 of 0.2856 and the lowestof 4.7624.The model was built by BP neural network method,with the highest~2 of 0.7212 and the lowest~2of 1.4912.The accuracy and stability of the model was greatly improved.