March 5, 2021

About the speaker

Dr. Weidong Kuang is from the Department of Electrical and Computer Engineering at UTRGV.

Deep Learning for Radar Automatic Target Recognition

Radar has long been used for military and non-military purposes in a wide variety of applications such as guidance, remote sensing, imaging and global positioning. Synthetic Aperture Radar (SAR) is a technique or a device which uses signal processing to improve the resolution beyond the limitation of physical antenna aperture. The target images generated by SAR carry abundant information about the target. However, the interpretation and understanding of SAR images are much different from optical photo analysis, and thus target recognition on SAR images by human eyes is challenging and impractical. Deep learning provides an efficient approach to automatic target recognition (ATR) for SAR applications.

In this talk, we will present our preliminary work on target classification using convolutional neural networks (CNNs) based on MSTAR dataset, and identify a few research directions. First, the principles of SAR and deep learning are briefly introduced. Second, we will describe the MSTAR dataset in details. Third, we apply three CNN architectures, called baselines, to target recognition on MSTAR dataset, and compare their performances. The results confirm that the classification accuracy strongly depends on the amount of training data. At the end, we will highlight the challenges of ATR for SAR and some ideas to tackle these challenges.