Bi-RDNet: Performance Enhancement for Remote Sensing Scene Classification with Rotational Duplicate Layers

Erdem Safa Akkul*, Berk Arıcan, Behçet Uğur Töreyin

*Corresponding author for this work

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

1 Citation (Scopus)

Abstract

We propose compact and effective network layer Rotational Duplicate Layer (RDLayer) that takes the place of regular convolution layer resulting up to 128 × in memory saving. Along with network accuracy, memory and power constraints affect design choices of computer vision tasks performed on resource-limited devices such as FPGAs (Field Programmable Gate Array). To overcome this limited availability, RDLayers are trained in a way that whole layer parameters are obtained from duplication and rotation of smaller learned kernel. Additionally, we speed up the forward pass via partial decompression methodology for data compressed with JPEG(Joint Photograpic Expert Group)2000. Our experiments on remote sensing scene classification showed that our network achieves ∼ 4 × reduction in model size in exchange of ∼ 4.5 % drop in accuracy, ∼ 27 × reduction with the cost of ∼ 10 % drop in accuracy, along with ∼ 2.6 × faster evaluation time on test samples.

Original languageEnglish
Title of host publicationAdvances in Computational Collective Intelligence - 13th International Conference, ICCCI 2021, Proceedings
EditorsKrystian Wojtkiewicz, Jan Treur, Elias Pimenidis, Marcin Maleszka
PublisherSpringer Science and Business Media Deutschland GmbH
Pages669-678
Number of pages10
ISBN (Print)9783030881122
DOIs
Publication statusPublished - 2021
Event13th International Conference on Computational Collective Intelligence, ICCCI 2021 - Virtual, Online
Duration: 29 Sept 20211 Oct 2021

Publication series

NameCommunications in Computer and Information Science
Volume1463
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937

Conference

Conference13th International Conference on Computational Collective Intelligence, ICCCI 2021
CityVirtual, Online
Period29/09/211/10/21

Bibliographical note

Publisher Copyright:
© 2021, Springer Nature Switzerland AG.

Keywords

  • 1-bit DCNN
  • Compressed domain
  • Remote sensing
  • Rotational duplicate layer
  • Scene classification

Fingerprint

Dive into the research topics of 'Bi-RDNet: Performance Enhancement for Remote Sensing Scene Classification with Rotational Duplicate Layers'. Together they form a unique fingerprint.

Cite this