CAEVT: Convolutional Autoencoder Meets Lightweight Vision Transformer for Hyperspectral Image Classification.

Area of Science:

• Remote Sensing
• Computer Vision
• Machine Learning

Background:

• Convolutional Neural Networks (CNNs) excel at local feature extraction in hyperspectral image (HSI) processing but suffer from heavyweight models and slow processing.
• Existing dimensionality reduction methods often lack the ability to capture complex nonlinear relationships within spectral bands.

Purpose of the Study:

• To develop a lightweight and efficient network for hyperspectral image classification.
• To simultaneously extract local and global information for improved classification accuracy.
• To address the limitations of traditional methods in capturing nonlinear spectral characteristics.

Main Methods:

• A lightweight vision transformer was designed to learn long-range dependencies and extract both local and global information.
• A three-dimensional convolutional autoencoder was employed to capture nonlinear characteristics between spectral bands.
• These components were integrated into a novel HSI classification network named CAEVT.

Main Results:

• The proposed CAEVT network demonstrated superior performance across four hyperspectral datasets.
• CAEVT showed particular effectiveness in scenarios with limited labeled samples.
• The network achieved high accuracy and efficiency in hyperspectral image classification tasks.

Conclusions:

• The CAEVT network offers an effective and efficient solution for hyperspectral image classification.
• The integration of convolutional autoencoders and lightweight vision transformers overcomes limitations of traditional CNNs.
• The approach shows significant promise for applications requiring accurate HSI classification with minimal labeled data.