Exploring Transformer-Based Approaches for Hyperspectral Image Classification: A Comparative Analysis

Abstract

Hyperspectral imaging (HSI) is a technique that captures and analyzes multiple spectral bands for each pixel in an image. Convolutional neural networks (CNNs) effectively extract local features for HSI classification but struggle with capturing sequential spectral signatures. To address this, we conducted a comparative study to assess the performance of transformers, which can handle sequential information, in HSI classification. We reviewed a few models in the literature, analyzing their architecture and performance. Using the Indian Pines dataset, we evaluated models like ViT, SSRN-ViT, SpectralFormer, and Double-ViT. Our results show that Double-ViT models outperform transformers for HSI classification. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.