Research Of Hyperspectral Image Classification Based On Deep Learning

Hyperspectral image remote sensing technology can simultaneously acquire spatial and spectral information of the target region by combining spectroscopy and imaging technology.Hyperspectral image contains abundant feature information,which makes it has great application value in fine classification of ground objects.Hyperspectral image classification(HSIC)has become a hot research direction in the field of earth observation.Recently,deep learning has been used for HSIC due to its powerful feature learning and classification ability.In view of this,this paper proposes a new method based on deep learning to extract and classify hyperspectral image features.Firstly,according to the idea that HSIC is similar to the semantic segmentation problem in computer vision,this paper adopts the DeepLab which has an excellent performance in semantic segmentation to excavate spatial features of the hyperspectral image(HSI)pixel to pixel.It breaks through the limitation of patch-wise feature learning in the most of existing deep learning methods used in HSIC.More importantly,it can extract features at multiple scales and effectively avoid the reduction of spatial resolution.By consulting the literature,it is the first time to apply DeepLab to deal with the HSI features extraction.Secondly,the individual spectral or spatial features can’t fully utilize the advantages of HSI,this paper proposes a HSIC framework based on the spatial-spectral features.The spatial features extracted by DeepLab and the spectral features are fused by a weighted fusion method,then the fused features are input into support vector machine for final classification.Finally,in order to evaluate the performance of the proposed feature extraction method and classification framework,this paper conducted experiments on two general HSI data sets(the Indian Pines data set and the University of Pavia data set).The experimental results demonstrate that the proposed framework outperformed the traditional methods and the existing deep learning-based methods,especially for small-scale classes.By designing comparative experiments and visualizing the fusion features,it is proved that the spatial features extracted by DeepLab really contribute to HSIC.