The Study Of High-temperature Targets Multisperal Identification Index Construction And Decision Tree Discrision Method

Common forms of surface high temperaturet targets include forest fires,prairie fires,coal seam spontaneous combustion,indigenous coking,oil well torches,volcanic eruptions and the like.Remote sensing identification of high temperature target and its attribute inversion have important theoretical and practical values for environmental monitoring,disaster warning and resource investigation.In general,the temperature of the high-temperature target?500 K+?is significantly higher than that of the surface?300 K?,and its apparent reflectivity is also higher than normal temperature in the shortwave infrared band?1.30³.00?m?.But in the Photographic infrared band?0.76¹.30?m?,it is generally lower.Compared with the mid-infrared and thermal infrared data,the high-temperature target could be identified more accurately and locate accurately in the shortwave infrared band,so it is regarded as the favorable band for high-temperature target recognition.At present,the methods of shortwave infrared high temperature target recognition mainly include normalized fire index?NDFI?,Mahalanobis multiple cut-off,Mahalanobis multiple classification,factor analysis and so on.Based on the ASD hyperspectral measurement datas and Landsat8 OLI multi-spectral remote sensing image datas and in-depth study of the spectral characteristics of typical ambient temperature and high temperature targets,this paper used the analysis of variance to construct separability metrics for feature band screening between high-temperature targets and various types of ambient temperature,identified the effective bands for high temperature target recognition,and constructed and screened the high temperature target optimal identification index.Focusing on the confusion between the high-temperature targets and the Color-steel roof buildings in the recognition results,the research focused on the decision tree classification method and established a decision tree model to achieve high-temperature target accuracy.The main research results are as follows:1.In the visible and photographic infrared bands,the differences in spectral characteristics between high-temperature targets and typical ambient temperature objects are small,but in the shortwave infrared?1.70 to 2.40?m wavelength range of ASD measured datas or the 6th and 7th bands of Landsat8 OLI remote sensing image datas?,the spectral characteristics of high-temperature objects and normal-temperature objects are significantly different,and their apparent reflectance is much higher than that of various typical normal-temperature objects.2.Based on the statistical data of training samples and the idea of variance analysis,the separability measure index constructed by the inter-group deviation/intra-group dispersion is used to identify the characteristic band of high temperature target.The results show that the favorable band sort of high temperature target recognition of Landsat8 OLI is the 7^th,5^th,and 4^th?SWIR?NIR?Red?bands.3.Based on several effective bands identified by high temperature targets,multiple spectral indexes in the form of difference,ratio,and ratio of difference combinationsareconstructed,andtheoptimalspectrumindex?B7-B5-B4?/?B7+B5+B4?for high temperature target identification is determined by comprehensive screening.The index distinguishes high-temperature targets between water bodies,burned-out sites,forest lands?sunny slopes?and all normal-temperature features best.For bare land,residential areas and forest lands?shady slopes?,the index is not optimal,but it is separable and the metrics are also large,so it's also valid index.Using this index to identify high-temperature targets,the results showed that the optimal spectral index can effectively distinguish high-temperature targets from all normal-temperature features except the Color-steel roof buildings,with an accuracy of79.24%.4.According to the spectral index recognition results,the high-temperature targest can be accurately identified by constructing a decision tree model that can effectively divide every two similar ground features.The recognition results showed that the decision tree classification method can effectively distinguish high temperature targets from various types of normal temperature ground features,and the recognition accuracy is 97.67%.It is superior to the spectral index method in classification accuracy,especially in distinguishing high temperature targets from similar color-steel roof structures,the classification is better.