Application Research In Glacier Information Extraction Based On U-NET Model Of Deep Learning

Mountain glaciers are not only important climate indicators,but also the largest freshwater resources on the earth.Our country has the richest mountain glacier in the world.Since the 20 th century,under the background of increasingly warming climate,the global glaciers are facing a state of retreat and melting,which has aroused the world’s concern about it.Glacier facing retreat and melt,accompanied by a series of glacier jumps,ice collapses,glacier mudslides,ice lake collapse and other secondary glacier disasters,seriously are threatening the nearby and middle and lower reaches of the residents of the personal and property safety.Due to the high altitude of mountain glaciers,rugged terrain and harsh climatic conditions,it is not possible to conduct massive field surveys of large areas of glacier changes.Remote sensing has become an indispensable means to study the changes of glaciers because of its increasingly abundant data,strong availability and low price.How to deal with a large number of remote sensing image data and accurate acquisition of land information in remote sensing images is the key task of using remote sensing images.Deep learning can quickly identify remote sensing image targets,speed up the efficiency of remote sensing image processing,and process large amounts of data quickly and accurately.The release of ENVI Deep Learning Module has opened the door to the wide application of deep learning in the field of remote sensing imaging.In this paper,the research area of Yanlian Mountain,Karakoram Mountain Area and Cross Mountain Area is analyzed by means of data collection and data processing,and the glacier distribution characteristics of the study area are analyzed.Respectively,the glacier information extraction method of the study area is carried out by the ratio method,the snow cover index method,the neural network method and the maximum similar method of the traditional supervised classification method,the ENVI deep learning method and so on.The main research content of this article is summarized as follows:(1)The traditional glacier extraction methods such as the ratio method,the snow cover index method,the neural network method and the maximum similar method of the traditional supervised classification method are applying on the research area.Thearticle will conclude the principe、application and question of those means.(2)The application of remote sensing image glacier extraction in the research area was carried out by ENVI Deep Learning method.Based on the training model parameters and the construction of ROI sample set,a series of control experiments are established to explore the best parameters of ENVI Deep Learning in the automatic extraction of glaciers and the method of ROI construction.The experiments prove the effectiveness of ENVI Deep Learning method in the extraction of glaciers.(3)The classification image obtained by ENVI Deep Learning method is compared with the remote sensing image obtained by the traditional glacier extraction method,and the advantages and disadvantages of ENVI deep learning method compared with other glacier extraction methods are obtained.In view of the problems existing in ENVI Deep Learning method,some measures and suggestions for improvement are put forward.