Research On Lithology Recognition Technology Of Aerial Hyperspectral Remote Sensing Based On Machine Learning

Airborne hyperspectral remote sensing has been widely used in geological exploration.In view of the various demands for lithologic mapping in the field of geological survey,remote sensing lithologic identification has gradually developed into an important research direction in remote sensing geology.However,there are still many problems in remote sensing lithology identification,such as poor performance in information processing,high requirement for prior knowledge of staff,and difficulty in large storage and processing of data.Therefore,it is urgent to seek a lithology identification processing and analysis method suitable for aerial hyperspectral remote sensing data.In recent years,machine learning,especially neural network,performed well in classification and recognition.With the strong support of the nuclear geological system for the study of new technologies and methods,this study applies the neural network method in machine learning to the field of remote sensing lithology identification,and takes Longshoushan area in Gansu Province as the research area to carry out the research of airborne hyperspectral remote sensing lithology identification based on machine learning.Lithology identification of airborne hyperspectral remote sensing based on machine learning mainly studied the following three aspects:?1?Design of the lithology identification process for airborne hyperspectral remote sensing based on machine learning.An airborne hyperspectral remote sensing lithology identification process is designed for remote sensing geology.?2?Dimension reduction and feature extraction of airborne hyperspectral remote sensing data based on auto-encoding neural network.The dimension reduction and feature extraction of the multi-band airborne hyperspectral remote sensing data were carried out by using the auto-encoding neural network method,and the experimental verification was carried out.?3?Airborne hyperspectral remote sensing lithology identification based on deep neural network method.The deep learning method is applied to the lithology classification,and the lithology identification based on the deep learning method is carried out and verified by experiments.The results show:?1?the airborne hyperspectral remote sensing lithology identification process based on machine learning meets the requirements of remote sensing lithology identification research,and perform well in remote sensing lithology identification according to the designed process,which is sufficient to provide reference for subsequent research.?2?On the experimental test data,the error of dimension reduction and feature extraction based on self-coding network is calculated to be0.0000334.The method has good effect and is feasible.?3?On the test data,the accuracy of the identification result is calculated as 87.1%;In the actual classification experiment,the classification results are compared with the geological map,and it is found that the classification results are somewhat similar to the actual geological map.On the whole,the identification effect of Q,Pt₂d³ and?₁ strata is good,while the effect of Pt₁^t,N₁ and O₃^3-2??is poor.In general,this method is feasible to some extent.