Land-cover Classification Of Multi-spectral LiDAR Data Using CNN With Optimized Hyper-parameters

Land cover monitoring is becoming more and more important for ecosystem protection,climate change research,land resource management and sustainable development.In the field of remote sensing,classification is a common method to obtain land cover information.Airborne multi-spectral LiDAR systems,which collect both spectral and surface geometrical data of ground objects,have become a fast,large-scale spatial data acquisition method.Multispectral LiDAR technology opens a new era due to the completeness and consistency of spectrum and spatial geometric data.In recent years,the convolutional neural network(CNN)has made significant breakthroughs in image classification,target recognition and natural language processing.Compared with traditional machine learning methods,it has more powerful feature learning and feature expression capabilities,and more stable robustness and fault tolerance.Therefore,this paper proposes a land cover classification method for multispectral LiDAR data using CNN.The aim of this paper is to establish a corresponding multispectral LiDAR data land cover model using CNN's feature extraction and data analysis capabilities,and to find the optimal hyper-parameters of the model through experiments,ultimately to improve the generalization ability and classification effect of the model.The main work and conclusions in this paper can be summarized as,(1)The CNN model is established,which reduces the model parameters only by stacking simple function layers to improve the classification efficiency.The control parameter method is used to analyze the parameter sensitivity of the model to find the optimal hyper-parameters.(2)The feasibility of CNN in multi-spectral LiDAR data land cover classification is verified by experiments.The advantages and disadvantages of the CNN model are also analyzed,with the comparison of common machine learning methods(Principal component analysis,PCA;Random forest,RF)and traditional deep learning methods(Alex Net;VGG16;Res Net50).Experimental results demonstrate that the multi-spectral LiDAR data and CNN model provide a promising solution to land use and land cover applications.