Research On Landslide Detection Algorithm Based On Deep Learning

With powerful ability to feature extraction and feature combination,deep learning(DL)can represent the complex nonlinear relationship between large-scale data.So DL is one of the hot topics in computer science.Although,the continuous improvement of computer performance gives us a good chance to increase the network layers of DL,it also pose challenges to construct the basic model and to train the parameters.Based on searching the key problem of constructing model and training parameters,this paper combine the characteristic of hyper-spectral remote sensing and landslide detection to analyze features of DL model.And,the determination method of objective function and optimization algorithm is analyzed in this paper detailedly.Finally,based on the characteristics of hyper-spectral remote sensing data and the feature of landslide targets,paper have completed the application of deep learning in the field of geological disasters.The main research contents and research features of this paper are as follows:(1)We further research on the construction of deep learning algorithm and its physical meaning.Based on the idea of feature driven classification,paper discuss the mapping ability of deep learning algorithm extracting the nonlinear characteristics of data.Then,the feature aggregation advantages of different network models in different data sets and the strategy of model merging are analyzed.(2)The influence of the optimization algorithm on the stability of the depth learning model is analyze.Combined with the deduction process of the conventional optimization algorithm,the characteristics of the depth learning optimization algorithm are also analyzed.Then,paper conclude the solutions of classification to the large-scale high dimensional sample data.(3)Based on the characteristics of hyper-spectral remote sensing image data and the image characteristics of landslides,paper studied and analyzed the selection and construction method of deep learning basic network.Then,a deep learning model for landslide detection is built.(4)Through the application of concrete examples and process analysis,we can obtain the optimal parameters for deep learning of landslide detection.Which acquire the mapping efficiency changes from conventional shallow image features to highdimensional deep abstract feature spaces in different parameters,and the parameter-tofeature combination.The effect of the path provides the basis for parameter optimization for deep learning to extract data ambiguity and recessive features.(5)This paper propose a spectral surface feature extraction model of landslide hazard.This model considers the internal material of landslides,distribution of debris,weathering degree and coupling relationship between the stable characteristics of landslide object image based on object-oriented.Then,mainly exploring an application method of spectral surface feature combination,which is under the single band and multiband.