Defence Science and Technology Organisation

About: Defence Science and Technology Organisation is a nonprofit organization based out in Canberra, Australian Capital Territory, Australia. It is known for research contribution in the topics: Radar & Clutter. The organization has 2465 authors who have published 3856 publications receiving 90614 citations.

Understanding Data Augmentation for Classification: When to Warp?

Abstract: In this paper we investigate the benefit of augmenting data with synthetically created samples when training a machine learning classifier. Two approaches for creating additional training samples are data warping, which generates additional samples through transformations applied in the data-space, and synthetic over-sampling, which creates additional samples in feature-space. We experimentally evaluate the benefits of data augmentation for a convolutional backpropagation-trained neural network, a convolutional support vector machine and a convolutional extreme learning machine classifier, using the standard MNIST handwritten digit dataset. We found that while it is possible to perform generic augmentation in feature-space, if plausible transforms for the data are known then augmentation in data-space provides a greater benefit for improving performance and reducing overfitting.

A synthetic time-reversal imaging method for structural health monitoring

Abstract: This paper presents an experimental and theoretical investigation of the applicability of the time-reversal concept to guided waves in plate-like structures, where the stress waves are dispersive and of multi-modes. It is shown that temporal and spatial focusing can be achieved through time reversal, although the dispersive behaviour of the flexural waves renders it impossible to exactly reconstruct the waveform of the original excitation. Based on the principle of the time-reversal concept, a digital imaging method suitable for distributed sensor/actuator networks has been developed. This new method, which overcomes the limitation of the conventional phased array method that operates under pulse-echo mode, provides an efficient imaging method for locating and approximate sizing of structural damages. In addition, it has been shown that signal strengths can be considerably enhanced by applying the present synthetic time-reversal method, thus reducing the number of sensors and actuators required to achieve a given signal-to-noise ratio.