Publications

Extracting the Features of the Shallowly Buried Objects using LeNet Convolutional Network

Elsaadouny, M. and Barowski, J. and Rolfes, I.

14TH EUROPEAN CONFERENCE ON ANTENNAS AND PROPAGATION, EUCAP 2020
Volume: Pages:
DOI: 10.23919/EuCAP48036.2020.9135701
Published: 2020

Abstract
The convolutional neural networks are considered as the best artificial intelligence algorithms for image classification problems. Generally, a ConvNet requires a very large number of images to be trained well and to achieve the best results. This paper investigates the implementation of the LeNet-5 convolutional network (ConvNet) for images classification using a small dataset. The dataset of interest compromises images of buried objects obtained by a ground penetrating radar (GPR), which is considered as an efficient tool for detecting and defining buried objects. One of the main problems facing this classification task is the limited available data. In deep learning algorithms, the ConvNet is usually trained using very large datasets, therefore the transfer learning has to be employed, as it is considered as a very important tool in deep learning when dealing with limited datasets. The LeNet-5 has been deployed and trained on the Fashion-MNIST dataset, and the learned features have been transferred to our GPR dataset. The network performance has been monitored and the classification results show a high degree of precision and accuracy. © 2020 EurAAP.

« back