Unlocking the Potential of Unlabeled Data: Semi-Supervised Learning for Stratification of Hyperspectral Images

Abstract

Hyperspectral imaging, a critical technology for spectral data acquisition, plays a pivotal role in various fields, including agriculture, medicine, defense, remote sensing and environmental monitoring. Nonetheless, an enduring obstacle in the domain of hyperspectral image analysis revolves around the restricted accessibility of annotated data, posing an impediment to the creation of precise classification models. In response to this challenge, we present an innovative semi-supervised framework that harmoniously combines unsupervised feature learning with the employment of graph-based convolutional neural networks (GCNs). Our approach harnesses the latent knowledge hidden within vast pools of unlabeled hyperspectral data using autoencoders, which extract meaningful features. These features are then incorporated into a GCN-based architecture, leveraging spatial relationships among neighboring pixels. The fusion of unsupervised autoencoder based learning and graph-based techniques enables our model to achieve remarkable classification accuracy, even in scenarios with minimal labeled samples. Through extensive experimentation, we demonstrate the superior performance and robustness of our methodology across a spectrum of hyperspectral imaging datasets. This work is a significant step in the realm of semi-supervised hyperspectral image analysis, unlocking the potential of unlabeled data and empowering accurate classification in data-scarce environments.

Topics & Concepts