Self-Supervised Assisted Semi-Supervised Residual Network for Hyperspectral Image Classification

Due to the scarcity and high cost of labeled hyperspectral image (HSI) samples, many deep learning methods driven by massive data cannot achieve intended expectations. Semi-supervised self-supervised algorithms have advantages in coping with this phenomenon. This paper primarily concentrates on applying strategies make strides semi-supervised HSI classification. Notably, we design an effective a unified assisted residual network (SSRNet) framework for The SSRNet contains two branches, i.e., branch. branch improves performance introducing perturbation via spectral feature shift. characterizes auxiliary tasks, including masked bands reconstruction order forecast, memorize discriminative features HSI. can better explore unlabeled samples improve classification performance. Extensive experiments four benchmarks datasets, Indian Pines, Pavia University, Salinas, Houston2013, yield average overall accuracy 81.65%, 89.38%, 93.47% 83.93%, which sufficiently demonstrate that exceed expectations compared state-of-the-art methods.