A Mutual Guidance Attention-Based Multi-Level Fusion Network for Hyperspectral and LiDAR Classification

Hyperspectral image (HSI) and light detection and ranging (LiDAR) data classification has attracted more and more attention in remote sensing. Convolution neural network (CNN) has been proven to be effective for HSI and LiDAR data classification. In this letter, a novel three-branch CNN is designed...

Full description

Saved in:
Bibliographic Details
Published in:IEEE geoscience and remote sensing letters 2022, Vol.19, p.1-5
Main Authors: Zhang, Tongzhen, Xiao, Song, Dong, Wenqian, Qu, Jiahui, Yang, Yufei
Format: Article
Language:English
Subjects:
Artificial neural networks
Classification
Convolution
convolutional neural network
Feature extraction
feature fusion
Fuses
Heating systems
hyperspectral image (HIS)
Hyperspectral imaging
Image classification
Information flow
Laser radar
Lidar
Modules
Neural networks
Remote sensing
Support vector machines
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Hyperspectral image (HSI) and light detection and ranging (LiDAR) data classification has attracted more and more attention in remote sensing. Convolution neural network (CNN) has been proven to be effective for HSI and LiDAR data classification. In this letter, a novel three-branch CNN is designed to learn spectral, spatial, and elevation features, each of which adopts the multi-level feature fusion (MLF) module to fuse the shallow and deep features. Furthermore, in order to fully fuse the spatial and elevation information, we propose a mutual guidance attention (MGA) module. The MGA module increases the information flow between spatial and elevation branches, highlights the features of interest, and weakens useless features. The proposed method is evaluated on public datasets Houston and Trento. Experimental results demonstrate that our proposed method can provide higher classification accuracy than some existing methods.
ISSN:1545-598X
1558-0571
DOI:10.1109/LGRS.2021.3112773