Data Augmentation and Spatial-Spectral Residual Framework for Hyperspectral Image Classification Using Limited Samples

Lin Zhou, Jinbiao Zhu, Jihao Yang, Jie Geng

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

2 Scopus citations

Abstract

Hyperspectral image classification is a prominent topic in many remote sensing applications, but the limited number of manually annotated samples leads to performance bottlenecks. To resolve this issue, a data augmentation and spatial-spectral residual framework is proposed for hyperspectral image classification using limited samples. Firstly, an unsupervised pseudo-sample generation method is proposed to augment the sample set, and the generalization capability of the model is improved by mixup operations. Then, to adequately extract the spatial-spectral features of hyperspectral images, a spatial-spectral residual framework is designed to improve the classification performance of the model. The qualitative and quantitative experiments were carried out on Indian Pines dataset to validate the effectiveness of the model.

Original languageEnglish
Title of host publicationProceedings of 2022 IEEE International Conference on Unmanned Systems, ICUS 2022
EditorsRong Song
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages490-495
Number of pages6
ISBN (Electronic)9781665484565
DOIs
StatePublished - 2022
Event2022 IEEE International Conference on Unmanned Systems, ICUS 2022 - Guangzhou, China
Duration: 28 Oct 202230 Oct 2022

Publication series

NameProceedings of 2022 IEEE International Conference on Unmanned Systems, ICUS 2022

Conference

Conference2022 IEEE International Conference on Unmanned Systems, ICUS 2022
Country/TerritoryChina
CityGuangzhou
Period28/10/2230/10/22

Keywords

  • data augmentation
  • deep learning
  • few sample classification
  • hyperspectral image

Fingerprint

Dive into the research topics of 'Data Augmentation and Spatial-Spectral Residual Framework for Hyperspectral Image Classification Using Limited Samples'. Together they form a unique fingerprint.

Cite this