Semi-Active Convolutional Neural Networks for Hyperspectral Image Classification
Jing Yao,Xiangyong Cao,Danfeng Hong,Xin Wu,Deyu Meng,Jocelyn Chanussot,Zongben Xu
DOI: https://doi.org/10.1109/TGRS.2022.3206208
IF: 8.2
2022-01-01
IEEE Transactions on Geoscience and Remote Sensing
Abstract:Owing to the powerful data representation ability of deep learning (DL) techniques, tremendous progress has been recently made in hyperspectral image (HSI) classification. Convolutional neural network (CNN), as a main part of the DL family, has been proven to be considerably effective to extract spatial-spectral features for HSIs. Nevertheless, its classification performance, to a great extent, depends on the quality and quantity of samples in the network training process. To select those samples, either labeled or unlabeled, which can be used to enhance the generalization ability of CNNs and further improve the classification accuracy, we propose an iterative semi-supervised CNNs framework by means of active learning and superpixel segmentation techniques, dubbed as semi-active CNNs (SA-CNNs), for HSI classification. More specifically, we start to pretrain a CNN-based model on a small-scale unbiased labeled set and infer unlabeled data using the trained model, i.e., generating pseudolabels. Then, the reliable samples, which consist of two parts: high label homogeneity and most informativeness, are actively selected from superpixel segments. These selected labeled and unlabeled samples with their labels and pseudolabels are refed into the next-round network training. Moreover, three different schedules, i.e., log-, exp-, and linear-schedules, are progressively adopted to fully explore their potentials in sample selection, until a labeling budget is finally reached. Extensive experiments are conducted on three benchmark HSI datasets, demonstrating substantial performance improvements of the proposed SA-CNNs over other similar competitors.