Study On The Automatic Extraction Method Of Lithological Information Based On WorldView-3 Data

Lithological mapping is an important application of remote sensing technology in geological field,and the key technology is the automatic extraction method of lithological information.With the successful launch and official operation of WorldView-3 satellite on August 13,2014,the WorldView-3 data with high spatial resolution and high radiation resolution brings new data sources to the high-precision lithological mapping,which has great geological application potential theoretically.But there is few related research.Meanwhile,whether some advanced classifier combination algorithms are suitable for the information identification and classification of this data requires further study.Therefore,the article selects Pobei area of the Xinjiang Uygur Autonomous Region,China as a study area,where bedrock outcrops are widely distributed.Based on the WorldView-3 data spectral information,an automatic extraction model of lithological information constructed by Random Forests(RF)algorithm was presented.Hereafter,the Semantic Texton Forests(STF)lithological information automatic extraction model was constructed by combining spectral and texture features.While mining the essential characteristics of data,the WorldView-3 data's lithological information automatic extraction methods are studied comprehensively.This paper mainly obtained the following research results:(1)Based on the World View-3 images,combined with field investigation data and regional geologic map,using image enhancement techniques,completed the visual interpretation of lithological map in the study area,which was the important reference basis of follow-up sample selection and accuracy verification,and it was also proved that the WorldView-3 image could be used for lithological visual interpretation.(2)According to the characteristics of WorldView-3 data and the lithological distribution in the study area,the RF and STF lithological information extraction models were constructed for the first time.The RF model performance was adjusted by the number of samples and decision trees.Meanwhile,from three aspects,which included the sample quantity,the number of decision trees and the scale size of segmentation function,the STF model was debugged.The mechanism of influencing factors was analyzed,and the model was optimized to achieve high precision lithological mapping application.The experimental results show that the classification accuracy of RF map is about 87.66%,and the classification accuracy of STF map is about 90.06%.(3)The lithological information extraction results of RF and STF models were comprehensively analyzed,and it revealed that:The high classification accuracy and ideal effect of RF and STF models are benefited from the high resolution of WorldView-3 data and the cooperative advantages of RF and STF algorithms,which are not sensitive to noise,faster speed and more stable,not easy to overfitting as well.Compared to the RF model,the advantages of the STF model mainly include three points: firstly,classification decision support information is more abundant when the spectral and texture features of the image are captured at the same time;secondly,the unique segmentation function guarantees classification accuracy and speed;thirdly,the scale size of the texton blocks can be adjusted according to the geology of the study area,and it is more universal.Therefore,the STF model can maximally meet the needs of high-precision lithological mapping.