Hyperspectral Image Spectral-Spatial Feature Extraction and Classification Using Vision Transformer

Author proposes a model to overcome the challenges in hyperspectral image classification, including high dimensionality of data, insufficient number of labelled samples and complexity. Author proposes Principal Component Analysis (PCA) for feature extraction and Vision Transformer (ViT) to find spectral features in data. The proposed model gives Overall Accuracy (OA) 91.38%, 92.99% and 93.91%, also testing time required is 40.33 seconds, 15.64 seconds and 14.59 seconds when provided training samples of $10 \%, 15 \%$ and 20% of Salina (SA) Dataset [1] respectively which shows Classification accuracy is nearly similar with reduction in training data. Testing time, however, decreases, showing a 63.8% reduction. This reduction indicates better generalization of the model, resulting in faster predictions during testing.