Hyperspectral Image Classification Based On Label Consistencies

Hyperspectral images(HSIs)contain high-resolution spectral information of land covers.In the past few decades,HSI classification has developed into a significant part of remote sensing.Deep learning approaches recently have been widely applied to the classification of HSI and achieve good capability.Deep learning can effectively extract features from HSI data comparing with other traditional handcrafted methods.Most deep learning methods extract the image features through traditional convolution,which has demonstrated impressive ability in HSI classification.However,traditional convolution can only operate convolutions with fixed size and weight on regular square image regions.Moreover,it refers to the spectral features of the adjacent pixels but ignores the spectral features of long-range data with the training sample.Although graph convolution network(GCN)can capture relationships based on the predefined graph that contains global information,the pixels' relationships for graph construction cannot be well ensured with limited iterations.Hence,the extracted features and image classification have the limited performance with the GCN.In response to the issues raised above,the main work of this article is described as following:Firstly,aiming to extract more representative and discriminative image features,in this paper,the deep feature learning with label consistencies(DFL-LC)method is developed to realize HSI classification.The structure of the proposed method contains a multi-scale convolutional neural network(MSCNN)and GCN.And the label consistency(LC)constraint is embedded in the objective function,and end-to-end optimization is implemented.Secondly,for obtaining the discriminative features,we add the label consistency of single pixels(LCSP)and label consistency of group pixels(LCGP)regularization in the objective function to boost the classification accuracy.LCSP ensures the LC between the outputs and the real label of the sample.LCGP refers to considering the long-range data and realizing label reuse to alleviate deficiently labelled samples problem.Finally,DFL-LC is optimized through an iterative algorithm for training and testing on the Indian Pines dataset,the Salinas dataset,and the University of Pavia dataset,and then compare it with other methods.The overall accuracy(OA),average accuracy(AA),and Kappa coefficient are used to evaluate the classification results.The influence of the parameters on the experimental results is discussed and analyzed.The experimental results fully demonstrate that DFL-LC is superior to other methods in both quantitative and qualitative aspects.