SAMPLE Dataset Objects Classification Using Deep Learning Algorithms

Loading...
Thumbnail Image
Date
2023-04
ORCID
Advisor
Referee
Mark
Journal Title
Journal ISSN
Volume Title
Publisher
Společnost pro radioelektronické inženýrství
Altmetrics
Abstract
The main topic of the article is automatic target classification of the synthetic aperture radar images based on the dataset composed of measured and synthetic data. The original contribution of the authors is their own topology of the convolutional neural network (CNN) with 1, 2, 3, and 4 tiers. The original convolutional neural network is used to classify radar images from the Synthetic And Measured Paired and Labeled Experiment (SAMPLE) dataset which consists of SAR imagery from publicly available datasets and well-matched synthetic data. The presented topologies of the CNN with 1, 2, 3, and 4 tiers were analyzed in 3 different scenarios: trained on the basis of real measured data and tested by synthetic data, trained on the basis of synthetic data, and tested by real measured data, and in the last case training and testing sets were formed by combining real measured and synthetic data. Based on the results of testing we could not use the proposed convolutional neural network trained with real measured data to classify synthetic radar images and vice versa (the 1st and the 2nd scenarios). The only last scenario with a combination of real measured and synthetic data in the training, validation, and testing data sets generates excellent results. The authors also present some confusion matrices, which can explain the reasons for the misclassification of radar images of military equipment. Comparing achieved results with another SAMPLE dataset classification results we can prove the usability of proposed and tested CNN structures for automatic target classification of the synthetic aperture radar images. The classification accuracy of the original convolutional network is 96.1%, which is better than the results of the other research so far.
Description
Citation
Radioengineering. 2023 vol. 32, č. 1, s. 63-73. ISSN 1210-2512
https://www.radioeng.cz/fulltexts/2023/23_01_0063_0073.pdf
Document type
Peer-reviewed
Document version
Published version
Date of access to the full text
Language of document
en
Study field
Comittee
Date of acceptance
Defence
Result of defence
Document licence
Creative Commons Attribution 4.0 International license
http://creativecommons.org/licenses/by/4.0/
Collections
Citace PRO