| The early warning of coal gangue dump reignition is the key and difficult point of the ecological safety work in the mining area.The study on the spectral response of surface vegetation under the reignition heat stress is an effective way to realize the early warning of coal gangue dump reignition.It is of great significance for the prevention and control of coal gangue dump reignition.In recent years,the existing acid coal gangue dumps have been effectively controlled nationwide,but the phenomenon of reignition still exists,and the large-scale surface vegetation coverage hinders the early warning and monitoring of coal gangue dumps by thermal infrared and other remote sensing technologies.The interior of the coal gangue dump is in the "self-heating" stage before the surface fire occurs,and the changes in the physiological characteristics of the surface plants under this reignition heat stress lead to specific changes in their spectral responses.Based on this,this thesis took alfalfa,a commonly used herbaceous vegetation for gangue dump control,as the research object,took the remote sensing monitoring of plant stress as the theoretical basis,simulated and explored the spectral response characteristics of alfalfa under reignition heat stress based on multi-source and multi-scale remote sensing data.The pot experiment was carried out to simulate the reignition heat stress,the identification of heat stress and its level were studied from two aspects:the direct response of the alfalfa canopy spectrum and the indirect spectral response of the physiochemical parameters;and the classification model of alfalfa reignition early warning levels was constructed by using multi-dimensional features.The main contents of the thesis are as follows:(1)In order to explore the characteristic spectral response caused by changes in the physiological state of alfalfa under reignition heat stress,the partial data of alfalfa in the LOPEX93 dataset was used.Spectral responses to changes in water and chlorophyll content of the alfalfa at leaf and canopy scales were simulated and analyzed based on the Prospect leaf radiative transfer model and the Sail canopy radiative transfer model.The simulated range of the leaf spectrum was 400nm-2400nm and that of the canopy spectrum was 400nm-1800nm.The results show that:① The leaf spectral response caused by changes in water content of alfalfa is mainly around 970nm,1200nm,1450nm and 1950nm;the canopy spectral reflectance decreases with the increase of the water content in the visible light region,and is positively correlated with the water content after the red edge.②The leaf spectral response caused by changes in chlorophyll content is mainly in the visible light region,especially at 500-650 nm,and the leaf spectral reflectance decreases significantly with the increase of chlorophyll content;the canopy spectral response is mainly in the visible light and red edge regions,the former is negatively correlated with the chlorophyll content,and the latter is reflected in the change of position and slope.(2)Aiming at the problem of identifying reignition heat stress and level discrimination by direct spectral response characteristics of canopy,the UAV multi spectral imagery and non-imaging canopy hyperspectral data of the potted alfalfa were used.Based on machine learning classification algorithm,vegetation index,JM distance,normalized mean distance,spectral derivative difference entropy and harmonic analysis,the spectral response characteristics were analyzed,and the direct monitoring of alfalfa under reignition heat stress was realized.① In the study of multispectral imagery,the image classification accuracy of k-nearest neighbor is the highest,with an overall accuracy of 94.84%and a Kappa coefficient of 0.93;the red-edge chlorophyll index(CIred-edge)has a strong ability to identify normal alfalfa and alfalfa under reignition heat stress.②In the study of canopy hyperspectral data,the time series analysis of hyperspectral index verifies the ability of CIred-edge to identify reignition heat stress;the red light absorption valley RW(640-680nm),red-edge RE(670-737nm)and near-infrared region NIR(780-900nm)are suitable as the preferred band;in the whole cycle,the variation law of the mean value of spectral derivative difference entropy in the RE band is obvious,but its application as a discriminant feature of the reignition heat stress level is relatively limited;in a single date,the harmonic sub-signal amplitude C1 in the RE band can be used as a frequency-domain feature to discriminate the reignition heat stress level.The results show that the vegetation index has the potential to extract the boundary of reignition heat stress,and the harmonic sub-signal amplitude C1 has the potential to map the areas of different heat stress levels under the surface vegetation coverage.(3)In order to study the indirect spectral response characteristics of water content and chlorophyll fluorescence parameters of alfalfa to discriminate the heat stress level,the vegetation index,correlation coefficient,Lasso regression and long short-term memory(LSTM)were used at leaf and canopy scales.Time series analysis and feature screening were used to select the best spectral features of suitable moisture indexes and fluorescence parameters to build SF-LSTM model,so as to discriminate the reignition heat stress of alfalfa.The results are as follows:①LFMC is the optimal water content index.Among its leaf spectral features,the verification set of SF-LSTM reignition heat stress discriminant model constructed by FDS(1661),RVI(1525,1771),DVI(1412,740)and NDVI(1447,1803)has the highest accuracy(around 90%)when the classification strategy is 3(the control group,60℃ and 90℃,120℃,150℃ and 180℃)and the time series length is 5(Five consecutive monitoring periods);among its canopy spectral features,the verification set of SF-LSTM reignition heat stress discriminant model constructed by RVI(599,611),DVI(570,634)and NDVI(652,671)has the highest accuracy(around 86%)when the classification strategy is 3 and the time series length is 3.②Fv’/Fm’ is the optimal chlorophyll fluorescence parameter.Among its leaf spectral features,the verification set of SF-LSTM reignition heat stress discriminant model constructed by FDS(510),RVI(567,698)and NDVI(553,701)has the highest accuracy(around 85%)when the classification strategy is 3 and the time series length is 3;among its canopy spectral features,the verification set of SF-LSTM reignition heat stress discriminant model constructed by RS(741),FDS(516)and NDVI(527,669)has the highest accuracy(around 87%)when the classification strategy is 3 and the time series length is 3.The results show that the reignition heat stress levels are suitably divided into three categories,the control group,60℃ and 90℃,120℃,150℃ and 180℃.(4)The discrimination results of the reignition heat stress level are consistent with previous research on the temperature distribution law of the spontaneous combustion coal gangue dump.The reignition early warning can be divided into three levels according to the risk level of surface ignition:Level 0(normal state,i.e.the control group),Level 1(minor dangerous,i.e.60℃ and 90℃),Level 2(dangerous,i.e.120℃,150℃ and 180℃).In order to construct a canopy-scale hierarchical classification model of reignition warning in alfalfa with high accuracy and strong generalization ability,the direct response characteristics of canopy harmonic analysis and indirect spectral response characteristics of physiochemical parameters were used as synergistic input data:① The reconstructed independent component analysis(RICA)algorithm can be used to realize the fusion and independent component extraction of mixed features in frequency domain and spectral dimension.②The discriminant model constructed by combining harmonic sub-signal amplitude,moisture and fluorescence spectral features has the highest classification accuracy.The accuracy of the validation set is about 95%.By t-SNE algorithm,the boundaries between different warning level categories are obvious,and the classification effect is good. |
PDF Full Text Request