HyperLiteNet: Extremely Lightweight Non-Deep Parallel Network for Hyperspectral Image Classification

Jianing Wang,Zhao Pei,Bo Liu,Siying Guo,Linhao Li,Runhu Huang

doi:10.3390/rs14040866

Jianing Wang, Zhao Pei + Show 4 more

Open Access

https://doi.org/10.3390/rs14040866

Copy DOI

Journal: Remote sensing	Publication Date: Feb 11, 2022
Citations: 3	License type: CC BY 4.0

Affiliation: Shaanxi Normal University, Xidian University

Abstract

Deep learning (DL) is widely applied in the field of hyperspectral image (HSI) classification and has proved to be an extremely promising research technique. However, the deployment of DL-based HSI classification algorithms in mobile and embedded vision applications tends to be limited by massive parameters, high memory costs, and the complex networks of DL models. In this article, we propose a novel, extremely lightweight, non-deep parallel network (HyperLiteNet) to address these issues. Based on the development trends of hardware devices, the proposed HyperLiteNet replaces the deep network by the parallel structure in terms of fewer sequential computations and lower latency. The parallel structure can extract and optimize the diverse and divergent spatial and spectral features independently. Meanwhile, an elaborately designed feature-interaction module is constructed to acquire and fuse generalized abstract spectral and spatial features in different parallel layers. The lightweight dynamic convolution further compresses the memory of the network to realize flexible spatial feature extraction. Experiments on several real HSI datasets confirm that the proposed HyperLiteNet can efficiently decrease the number of parameters and the execution time as well as achieve better classification performance compared to several recent state-of-the-art algorithms.

Highlights

With continuous improvement to the accuracy of image acquisition equipment, the spectral dimension of hyperspectral images (HSIs) has expanded greatly, which provides great discrimination ability for the component analysis of materials and land covers, etc
Traditional classifiers can be mainly divided into two categories, according to their feature extraction characteristics: (1) spectral-based classification algorithms, such as support vector machines (SVM) [10], K-nearest neighbor [11], multinomial logistic regression [12], and random forest [13]; (2) spectral–spatial-based classification algorithms, e.g., SVM
The proposed HyperLiteNet can be roughly divided into five main components: (1) the parallel interconnection module (PIM); (2) the pointwise convolution branch (PCB) to extract spectral features; (3) the dynamic convolution branch (DCB) to extract spatial features; (4) the feature interconnection module (FIM); (5) the classification module (CM)

Summary

Introduction

With continuous improvement to the accuracy of image acquisition equipment, the spectral dimension of hyperspectral images (HSIs) has expanded greatly, which provides great discrimination ability for the component analysis of materials and land covers, etc. Many linear and nonlinear dimension reduction algorithms, such as band selection [6], principal component analysis [7], independent component analysis [8], and the maximum noise fraction [9] have been proposed to address this problem. Such algorithms employ feature optimization to remove redundant features and improve classification performance. Traditional classifiers can be mainly divided into two categories, according to their feature extraction characteristics: (1) spectral-based classification algorithms, such as support vector machines (SVM) [10], K-nearest neighbor [11], multinomial logistic regression [12], and random forest [13]; (2) spectral–spatial-based classification algorithms, e.g., SVM

Methods

Findings

Discussion

Conclusion