Skip to main navigation Skip to search Skip to main content

A Novel Approach to Lighten the Onboard Hyperspectral Anomaly Detector

  • Ning Ma
  • , Yu Peng
  • , Shaojun Wang*
  • , Jingyi Dong
  • *Corresponding author for this work
  • Harbin Institute of Technology

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

Abstract

Hyperspectral image (HSI) anomaly targets detection is always applied for timeliness and onboard mission. For high detection accuracy, deep learning based HSI anomaly detectors (ADs) are widely employed in recent researches. However, their huge network scale for high-level representation ability leads to great computation burden for the onboard computation system. To decrease the computation complexity of the detector, a lightweight network is expected for the HSI AD. In this paper, by creating a multiobjective optimization with nondominated sorting genetic algorithm II (NSGA-II), an automatic evolution based deep learning network HSI AD (Auto-EDL-AD) is proposed to explore a lightweight network. The experimental results on an HSI dataset show that the proposed Auto-EDL-AD can generate an optimal network for the HSI anomaly detection which reaches up to 170% speedup without any detection accuracy loss.

Original languageEnglish
Title of host publicationWireless and Satellite Systems - 10th EAI International Conference, WiSATS 2019, Proceedings
EditorsMin Jia, Qing Guo, Weixiao Meng
PublisherSpringer Verlag
Pages432-445
Number of pages14
ISBN (Print)9783030191559
DOIs
StatePublished - 2019
Event10th EAI International Conference on Wireless and Satellite Systems, WiSATS 2019 - Harbin, China
Duration: 12 Jan 201913 Jan 2019

Publication series

NameLecture Notes of the Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering, LNICST
Volume281
ISSN (Print)1867-8211

Conference

Conference10th EAI International Conference on Wireless and Satellite Systems, WiSATS 2019
Country/TerritoryChina
CityHarbin
Period12/01/1913/01/19

Keywords

  • Deep learning
  • Hyperspectral image
  • Multiobjective optimization
  • Real-time processing

Fingerprint

Dive into the research topics of 'A Novel Approach to Lighten the Onboard Hyperspectral Anomaly Detector'. Together they form a unique fingerprint.

Cite this