Few-Shot Learning Based on Embedded Self-Distillation and Adaptive Wasserstein Distance for Hyperspectral Image Classification

Due to the domain shift, it is challenging to achieve ideal experimental results for cross-domain few-shot learning (FSL) in hyperspectral image (HSI) classification. Most existing FSL algorithms are impacted by the limited samples, and they do not effectively leverage the representations from different layers of the network. Therefore, this article proposes an FSL based on embedded self-distillation and adaptive Wasserstein (ESAW-FSL) distance for HSI classification. First, the embedding self-distillation network is proposed in the feature extraction process of the source domain (SD) and the target domain (TD). The embedding self-distillation network utilizes self-distillation from different perspectives to get discriminative features. In the SD, the mask evaluation of embedded features is employed to guarantee the learning of guiding features. Second, a domain adaptation based on adaptive Wasserstein distance is designed to alleviate the domain shift problem between the domains. A lightweight feature correlation network learns the comprehensive cost matrix in the Wasserstein distance adaptively, and the obtained cost matrix helps achieve domain adaptation by an iterative algorithm. Finally, a focal loss based on double softening is adopted in the process of FSL. The probability is double softened to improve the ratio of correctly classifying hard samples. Experiments are conducted on three widely used hyperspectral datasets and compared with six state-of-the-art algorithms. The overall accuracy (OA) and average accuracy (AA) are achieved in multiple experiments, demonstrating the effectiveness of ESAW-FSL.

I. Introduction

With the increasing development of artificial intelligence technology and onboard and airborne imaging spectrometers, hyperspectral image (HSI) classification technology has become booming and widespread [1]. In numerous remote sensing image applications [2], [3], the HSI has become a unique branch due to the rich spectral bands [4]. Its increasingly important role in agricultural production [5], land detection [6], geological research [7], ecological monitoring [8], and national defense maintenance [9] has attracted a growing number of researchers’ attention [10]. HSI classification technology analyzes the spatial details and spectral bands of an image to learn each pixel’s spatial and spectral features [11], thereby determining the class of each pixel. Despite rich spectral information, time-consuming and expensive labels constrain improving classification results [12].