PulseNetOne: Fast Unsupervised Pruning of Convolutional Neural Networks for Remote Sensing

Scene classification is an important aspect of image/video understanding and segmentation. However, remote-sensing scene classification is a challenging image recognition task, partly due to the limited training data, which causes deep-learning Convolutional Neural Networks (CNNs) to overfit. Anothe...

Full description

Bibliographic Details
Main Authors: David Browne, Michael Giering, Steven Prestwich
Format: Article
Language:English
Published: MDPI AG 2020-03-01
Series:Remote Sensing
Subjects:
Online Access:https://www.mdpi.com/2072-4292/12/7/1092
_version_ 1797626407122108416
author David Browne
Michael Giering
Steven Prestwich
author_facet David Browne
Michael Giering
Steven Prestwich
author_sort David Browne
collection DOAJ
description Scene classification is an important aspect of image/video understanding and segmentation. However, remote-sensing scene classification is a challenging image recognition task, partly due to the limited training data, which causes deep-learning Convolutional Neural Networks (CNNs) to overfit. Another difficulty is that images often have very different scales and orientation (viewing angle). Yet another is that the resulting networks may be very large, again making them prone to overfitting and unsuitable for deployment on memory- and energy-limited devices. We propose an efficient deep-learning approach to tackle these problems. We use transfer learning to compensate for the lack of data, and data augmentation to tackle varying scale and orientation. To reduce network size, we use a novel unsupervised learning approach based on k-means clustering, applied to all parts of the network: most network reduction methods use computationally expensive supervised learning methods, and apply only to the convolutional or fully connected layers, but not both. In experiments, we set new standards in classification accuracy on four remote-sensing and two scene-recognition image datasets.
first_indexed 2024-03-11T10:09:59Z
format Article
id doaj.art-87edacbb49d543c780225b985308f48c
institution Directory Open Access Journal
issn 2072-4292
language English
last_indexed 2024-03-11T10:09:59Z
publishDate 2020-03-01
publisher MDPI AG
record_format Article
series Remote Sensing
spelling doaj.art-87edacbb49d543c780225b985308f48c2023-11-16T14:34:33ZengMDPI AGRemote Sensing2072-42922020-03-01127109210.3390/rs12071092PulseNetOne: Fast Unsupervised Pruning of Convolutional Neural Networks for Remote SensingDavid Browne0Michael Giering1Steven Prestwich2Insight Centre for Data Analytics, University College Cork, Cork City T12 XF62, IrelandUnited Technologies Research Centre, Penrose Wharf Business Centre, Penrose Quay, Cork City T23 XN53, IrelandInsight Centre for Data Analytics, University College Cork, Cork City T12 XF62, IrelandScene classification is an important aspect of image/video understanding and segmentation. However, remote-sensing scene classification is a challenging image recognition task, partly due to the limited training data, which causes deep-learning Convolutional Neural Networks (CNNs) to overfit. Another difficulty is that images often have very different scales and orientation (viewing angle). Yet another is that the resulting networks may be very large, again making them prone to overfitting and unsuitable for deployment on memory- and energy-limited devices. We propose an efficient deep-learning approach to tackle these problems. We use transfer learning to compensate for the lack of data, and data augmentation to tackle varying scale and orientation. To reduce network size, we use a novel unsupervised learning approach based on k-means clustering, applied to all parts of the network: most network reduction methods use computationally expensive supervised learning methods, and apply only to the convolutional or fully connected layers, but not both. In experiments, we set new standards in classification accuracy on four remote-sensing and two scene-recognition image datasets.https://www.mdpi.com/2072-4292/12/7/1092pruning networksnetwork compressionremote-sensing image classificationtransfer learningConvolutional Neural Networkpre-trained AlexNet
spellingShingle David Browne
Michael Giering
Steven Prestwich
PulseNetOne: Fast Unsupervised Pruning of Convolutional Neural Networks for Remote Sensing
Remote Sensing
pruning networks
network compression
remote-sensing image classification
transfer learning
Convolutional Neural Network
pre-trained AlexNet
title PulseNetOne: Fast Unsupervised Pruning of Convolutional Neural Networks for Remote Sensing
title_full PulseNetOne: Fast Unsupervised Pruning of Convolutional Neural Networks for Remote Sensing
title_fullStr PulseNetOne: Fast Unsupervised Pruning of Convolutional Neural Networks for Remote Sensing
title_full_unstemmed PulseNetOne: Fast Unsupervised Pruning of Convolutional Neural Networks for Remote Sensing
title_short PulseNetOne: Fast Unsupervised Pruning of Convolutional Neural Networks for Remote Sensing
title_sort pulsenetone fast unsupervised pruning of convolutional neural networks for remote sensing
topic pruning networks
network compression
remote-sensing image classification
transfer learning
Convolutional Neural Network
pre-trained AlexNet
url https://www.mdpi.com/2072-4292/12/7/1092
work_keys_str_mv AT davidbrowne pulsenetonefastunsupervisedpruningofconvolutionalneuralnetworksforremotesensing
AT michaelgiering pulsenetonefastunsupervisedpruningofconvolutionalneuralnetworksforremotesensing
AT stevenprestwich pulsenetonefastunsupervisedpruningofconvolutionalneuralnetworksforremotesensing