Study on Quantitative Extraction of Coal SeamFire from Hyper Spectral Remote Sensing Information
|School||Capital Normal University|
|Course||Cartography and Geographic Information Systems|
|Keywords||Coal Seam Fire Hyper Spectral Remote Sensing Qualitative Extraction Spontaneous Information|
Coal seam fire, as one of the main hazards of coal, consumes a large amount of energy, releases a large quantity greenhouse gas, and causes heavy contamination of atmosphere, water and soil. As well as it threats residents’living and working nearby the fire zone. As a global concern, the coal seam fire management has been researched in many countries. In China, the problem is more serious, faced with the loss of natural resources and damage of ecological environment, related functional sectors bring to high attention and conduct research, monitoring and management.My thesis selected Wuda Coalfield in Inner Mongolia as the study area, according to the the distribution of coal fire in China. By the regional geological data and the pre-project field survey data, etc., based on the hyper spectral remote sensing methods, according to the underground coals reveal surface features of the landscape, my thesis researched to detect the spectrum change characteristics of rock and soil caused by coal seam fire, and to identify specific reflect of these changes in remote sensing images. The quantitative remote sensing inversion model established for detection the underground coal seam fire with satellite-based hyperspertal image, by using hyperion hyperspectral data, through the statistical software R. My thesis implents the technology to extraction the coal fire information combined with field monitoring data, and verifies the effectiveness and accuracy of the results. At last, it provids the actionable quantitative extraction methods and results as reference for the coal seam fire monitoring and environmental control.The paper studies two aspects: one side is about the basic data analysis and hyperspectral data pre-processing, including analysis for spectral characteristics and hyperion data pre-processing; the other side is about extraction of information, including the screening of diagnosis factors for information extraction model and the quantitative comparison and establishmen for model.The research according to satellite-borne high spectrum data, different from multi- spectrum and airborne high spectrum.The quantity inversion model establishment is based on the spectrum charactor of high spectrum image itself, with the selection of diagnosis factors on contribution rate. Then its result quantity compared and analyzed by ROC.During the research on the quantitative extraction of coal seam dire from hyper spectral remote sensing information, this paper achieved main results as follow:1) Obtain the samples for laboratory testing and detail analysis of spectrum, and determine the ground feature diagnostic characteristics in coalfield. Deep study the Hyperion hyper spectral remote sensing image preprossing, promote the image quality, especially in strip removaling, atmospheric correction and smooting, in order to reduce the interfering factors, and facilitate the follow-up information extraction and analysis.2) Screening of diagnosis factors on contribution rate for information extraction model, based on the model of choice theory, confusion matrix, as well as self-help law.3) Establish quantity inversion model based on machine learning, classification and prediction theory, by generalized linear model and generalized additive model, and then quantity compare and analyze by ROC curve. Complish the underground coal seam fire extrection map and predict its distribution.