Litcius/Paper detail

Effective hyperspectral image classification based on segmented PCA and 3D-2D CNN leveraging multibranch feature fusion

Masud Ibn Afjal, Md. Nazrul Islam Mondal, Md. Al Mamun

2024Journal of Spatial Science17 citationsDOI

Abstract

We present an innovative hyperspectral image (HSI) classification method addressing challenges posed by closely spaced wavelength bands. Our approach combines 3D-2D convolutional neural networks (CNNs) with multi-branch feature fusion for improved spectral-spatial feature extraction. Using segmented principal component analysis (Seg-PCA), we reduce HSIs’ spectral dimensions into global and local intrinsic characteristics. The integration of 3D and 2D CNNs captures joint spectral-spatial features, while a multi-branch network extracts and merges diverse local features along the spectral dimension. Our method outperforms existing approaches, achieving remarkable accuracy of 99.27%, 100%, and 99.99% on Indian Pines, Salinas Scene, and University of Pavia datasets, respectively.

Topics & Concepts

Hyperspectral imagingPattern recognition (psychology)Artificial intelligenceFeature (linguistics)FusionComputer scienceImage (mathematics)Image fusionComputer visionLinguisticsPhilosophyRemote-Sensing Image ClassificationAdvanced Image Fusion TechniquesRemote Sensing and Land Use
Effective hyperspectral image classification based on segmented PCA and 3D-2D CNN leveraging multibranch feature fusion | Litcius