Hyperspectral Image Classification Utilizing Convolutional Neural Network and Transformer

In hyperspectral image (HSI) classification, each pixel is assigned a land cover type. In recent years, convolutional neural network (CNN)-based HSI classification methods have significantly improved classification performance due to their exceptional feature extraction capabilities. However, these methods are limited in capturing deep semantic features, while Transformers excel at representing such features. This paper proposes a CNN combined with Transformer (CT) approach to extract spatial-spectral joint features of HSI for classification. Using three standard HSI datasets, experimental analysis demonstrates that the CT method achieves superior classification performance compared to other methods.