GAF-NAU: Gramian Angular Field encoded Neighborhood Attention U-Net for Pixel-Wise Hyperspectral Image Classification

• DOI: 10.48550/arxiv.2204.10099
• Publication Date: 2022-01-01
• Authors: Sidike Paheding; Abel A. Reyes; Anush Kasaragod; Thomas Oommen
• Keywords: Feature (linguistics); Feature vector

Hyperspectral image (HSI) classification is the most vibrant area of research in hyperspectral community due to rich spectral information contained HSI can greatly aid identifying objects interest. However, inherent non-linearity between materials and corresponding profiles brings two major challenges classification: interclass similarity intraclass variability. Many advanced deep learning methods have attempted address these issues from perspective a region/patch-based approach, instead pixel-based alternate. patch-based approaches hypothesize that neighborhood pixels target pixel fixed spatial window belong same class. And this assumption not always true. To problem, we herein propose new architecture, namely Gramian Angular Field encoded Neighborhood Attention U-Net (GAF-NAU), for classification. The proposed method does require regions or patches centered around raw perform 2D-CNN based classification, instead, our approach transforms 1D vector into 2D angular feature space using (GAF) then embed it attention network suppress irrelevant while emphasizing on pertinent features useful task. Evaluation results three publicly available datasets demonstrate superior performance model.