eprintid: 12233
rev_number: 2
eprint_status: archive
userid: 1
dir: disk0/00/01/22/33
datestamp: 2023-11-10 03:26:46
lastmod: 2023-11-10 03:26:46
status_changed: 2023-11-10 01:17:11
type: article
metadata_visibility: show
creators_name: Qayyum, A.
creators_name: Malik, A.
creators_name: M Saad, N.
creators_name: Mazher, M.
title: Designing deep CNN models based on sparse coding for aerial imagery: a deep-features reduction approach
ispublished: pub
keywords: Aerial photography; Antennas; Benchmarking; Convolution; Convolutional neural networks; Deep learning; Feature extraction; Remote sensing; Unmanned aerial vehicles (UAV), Classification algorithm; Convolutional networks; Dimension reduction; Features reductions; Performance outcome; Scene classification; Sparse coding; UAV remote sensing, Learning algorithms, artificial neural network; design method; image analysis; numerical method; principal component analysis; remote sensing; software; unmanned vehicle
note: cited By 13
abstract: Traditional methods focus on low-level handcrafted features representations and it is difficult to design a comprehensive classification algorithm for remote sensing scene classification problems. Recently, convolutional neural networks (CNNs) have obtained remarkable performance outcomes, setting several remote sensing benchmarks. Furthermore, direct applications of UAV remote sensing images that use deep convolutional networks are extremely challenging given high input data dimensionality with relatively small amounts of available labelled data. We, therefore, propose a CNN approach to scene classification that architecturally incorporates sparse coding (SC) technique for dimension reduction to minimize overfitting. Outcomes were compared with principal component analysis (PCA) and global average pooling (GAP) alternatives that use fully connected layer(s) in pre-trained CNN architecture(s) to minimize overfitting. SC was used to encode deep features extracted from the last convolutional layer of pre-trained CNN models by using different features maps in which deep features had been converted into low-dimensional SC features. These same sparse-coded features were concatenated by means of different pooling techniques to obtain global image features for scene classification. The proposed algorithm outperformed current state-of-the-art algorithms based on handcrafted features. When using our own UAV-based dataset and existing datasets, it was also exceptionally efficient computationally when learning data representations, producing a 93.64 accuracy rate. © 2019, © 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
date: 2019
publisher: Taylor and Francis Ltd.
official_url: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85062424629&doi=10.1080%2f22797254.2019.1581582&partnerID=40&md5=b6a255f6b0d92c6e5d6d39ab53ba17c3
id_number: 10.1080/22797254.2019.1581582
full_text_status: none
publication: European Journal of Remote Sensing
volume: 52
number: 1
pagerange: 221-239
refereed: TRUE
issn: 22797254
citation:   Qayyum, A. and Malik, A. and M Saad, N. and Mazher, M.  (2019) Designing deep CNN models based on sparse coding for aerial imagery: a deep-features reduction approach.  European Journal of Remote Sensing, 52 (1).  pp. 221-239.  ISSN 22797254