Abstract:In recent years, deep learning methods have achieved remarkable success in hyperspectral image classification (HSIC), and the utilization of convolutional neural networks (CNNs) has proven to be highly effective. However, there are still several critical issues that need to be addressed in the HSIC task, such as the lack of labeled training samples, which constrains the classification accuracy and generalization ability of CNNs. To address this problem, a deep multi-scale attention fusion network (DMAF-NET) is proposed in this paper. This network is based on multi-scale features and fully exploits the deep features of samples from multiple levels and different perspectives with an aim to enhance HSIC results using limited samples. The innovation of this article is mainly reflected in three aspects: Firstly, a novel baseline network for multi-scale feature extraction is designed with a pyramid structure and densely connected 3D octave convolutional network enabling the extraction of deep-level information from features at different granularities. Secondly, a multi-scale spatial–spectral attention module and a pyramidal multi-scale channel attention module are designed, respectively. This allows modeling of the comprehensive dependencies of coordinates and directions, local and global, in four dimensions. Finally, a multi-attention fusion module is designed to effectively combine feature mappings extracted from multiple branches. Extensive experiments on four popular datasets demonstrate that the proposed method can achieve high classification accuracy even with fewer labeled samples.

What problem does this paper attempt to address?

The paper primarily addresses the issue of limited labeled samples in the task of hyperspectral image classification (HSIC) by proposing a new solution. Specifically, the paper introduces a deep learning network architecture named DMAF-NET (Deep Multi-Scale Attention Fusion Network), which aims to improve the accuracy of hyperspectral image classification using limited training samples. The paper points out that although convolutional neural networks (CNNs) have achieved significant success in the field of hyperspectral image classification, the classification accuracy and generalization ability are affected when the number of training samples is limited. To solve this problem, DMAF-NET employs methods such as multi-scale feature extraction, spatial-spectral attention mechanism, and channel attention mechanism to enhance the learning ability with limited samples and improve classification performance. The main contributions of DMAF-NET include: 1. Designing a novel baseline network for multi-scale feature extraction, which combines a pyramid structure and densely connected 3D octave convolution network to extract deep information from different granularities. 2. Proposing a 3D multi-scale spatial-spectral attention module and a 4D pyramid multi-scale channel attention module to model the comprehensive dependencies between feature maps in four dimensions (spatial coordinates, direction, local, and global). 3. Designing a multi-attention feature fusion module that effectively integrates feature maps from different branches. 4. Experimental results on four popular hyperspectral datasets show that DMAF-NET can achieve high classification accuracy even under the condition of limited labeled samples. In summary, this study provides a valuable reference for related research by proposing a new method that can effectively perform hyperspectral image classification with limited samples.

DMAF-NET: Deep Multi-Scale Attention Fusion Network for Hyperspectral Image Classification with Limited Samples

Multiscale 3-D-2-D Mixed CNN and Lightweight Attention-Free Transformer for Hyperspectral and LiDAR Classification

Hyperspectral and Multispectral Image Fusion Based on Deep Attention Network.

Multiscale Feature Fusion Network Incorporating 3D Self-Attention for Hyperspectral Image Classification

A Feature Embedding Network with Multiscale Attention for Hyperspectral Image Classification

AMFAN: Adaptive Multiscale Feature Attention Network for Hyperspectral Image Classification

A Multiscale Dual-Branch Feature Fusion and Attention Network for Hyperspectral Images Classification

Subpixel Multilevel Scale Feature Learning and Adaptive Attention Constraint Fusion for Hyperspectral Image Classification

HAMNet: hyperspectral image classification based on hybrid neural network with attention mechanism and multi-scale feature fusion

Multiscale Information Fusion for Hyperspectral Image Classification Based on Hybrid 2D-3D CNN

Multiscale Densely Connected Attention Network for Hyperspectral Image Classification

Hyperspectral Image Classification Based on Dual-Scale Dense Network with Efficient Channel Attentional Feature Fusion

Multi-Scale Dense Networks for Hyperspectral Remote Sensing Image Classification

Attention Multihop Graph and Multiscale Convolutional Fusion Network for Hyperspectral Image Classification

An Effective Hyperspectral Image Classification Network Based on Multi-Head Self-Attention and Spectral-Coordinate Attention

Integrating Hybrid Pyramid Feature Fusion and Coordinate Attention for Effective Small Sample Hyperspectral Image Classification

A Cross-Channel Dense Connection and Multi-Scale Dual Aggregated Attention Network for Hyperspectral Image Classification

Multilevel Attention Dynamic-Scale Network for HSI and LiDAR Data Fusion Classification

Lightweight Multilevel Feature Fusion Network for Hyperspectral Image Classification

CMAAC: Combining Multiattention and Asymmetric Convolution Global Learning Framework for Hyperspectral Image Classification