Silicate Rocks Identification Using Remote Sensing Based On Thermal Radiation And Textural Features

With the popularization of application and research,remote sensing has been deeply integrated with petrology and the requirements of application are increasing,which needs lthological classification finer and more stable,and more rocks can be identified.The traditional study of remote sensing petrology is mainly focused on the spectroscopy field,in which,lithological identification is based on the reflection or emission features of rocks or minerals and methods for lithological identification is based on the spectrum analysis and image information enhancement.This paper explore the ability of new features in lithological identification from two aspects of rock thermal radiation characteristics and spatial texture feature and to build stable silicate rocks identification model and classification strategy,which provides a feasible solution for lithology fine identification.The study area is located in Qilian Mountain County,Qinghai province,which lies in northern Qinghai-Tibet Plateau.The study objects include mafic-ultramafic rock(which can be further fine divided into gabbro rock and serpentinitite rock),granite rock and quartz-rich rock.Landsat 8data were selected in this paper,including multi-temporal thermal infrared data sets and a single phase of visible-shortwave infrared data sets.Based on the theory of energy exchange between well-exposed silicate rocks and environmental system,the thermal characteristic index(??)that characterizes rockthermal radiation features was built.Based on the multi-temporal thermal infrared data sets,the effectiveness of ?? for lithological identification in differentenvironment was analyzed.The theoretical analysis shows that ?? has a sensitiverespond to lithological difference in simple environment system.However in remote sensing classification,the complex,variable and accidental factors in environment are unavoidable,which results in unstable results for lithological identification using a single ??.Considering the ?? in differentenvironment,a multi-temporal thermal feature space(MTTFS)model integrating multi-features was built to reduce the interference from environmental factors and to improve the accuracy and stability of lithological classification.The results show thatMTTFS model can effectively identify mafic-ultramafic rock and quartz-rich rock,and the identification accuracy of them are about 76%.In order to further realize lithology fine identification,spatial correction in neighborhood pixels was fully used.Fractal dimension describing rock surface textual information was used to study the ability of textural information in lithological identification.Further,Jeffries-Matusita(J-M)distance was used to evaluate the sensitivity of different textural bands and to select the optimal feature combination for lithology fine classification.It is found that different textual vectors have different sensitivity for lithological identification and more textural vectors yield higher class separability.The overall classification accuracy achieved by the optimal feature combination is improved by 9.04%,and the classification accuracy of the four rocks is all improved,especially for granite,showing an increase of 41.05%,which indicated the ability of spatial features in lithological identification and the importance and necessity of textural features evaluation.