Deep Learning for Weak Target Detection in Range-Doppler Data

Dang, Bibi

Deep Learning for Weak Target Detection in Range-Doppler Data

dc.contributor.advisor	Hougen, Dean
dc.contributor.author	Dang, Bibi
dc.contributor.committeeMember	Metcalf, Justin
dc.contributor.committeeMember	Diochnos, Dimitrios
dc.date.accessioned	2022-05-05T21:35:40Z
dc.date.available	2022-05-05T21:35:40Z
dc.date.issued	2022-05
dc.date.manuscript	2022-05
dc.description.abstract	A consistent issue for detectors in radar systems is how to correctly distinguish target signals from random noise. This is especially true for weak targets with low signal-to-noise ratios (SNRs). Traditional target detection techniques, such as constant false alarm rate (CFAR) detectors, apply detection thresholds that must be set to maximize the probability of detection (PD) while minimizing the probability of false alarm (PFA). These traditional detection techniques also struggle with increasing levels of computational complexity in low-SNR environments. This work investigates the application of deep neural networks towards the radar target detection problem. Two neural network architectures, NoisyLSTM and U-Net, are tested on range-Doppler data to identify regions of interest in which targets may be present. The U-Net model demonstrates promising results, producing detection predictions with a PD of 0.97 and PFA of 0.01 for targets captured by a staring radar at 10dB input SNR. This deep learning architecture may serve as a valuable preprocessing step to reduce the search space of more sophisticated radar detectors.	en_US
dc.identifier.uri	https://hdl.handle.net/11244/335555
dc.language	en_US	en_US
dc.subject	Machine Learning	en_US
dc.subject	Deep Learning	en_US
dc.subject	Radar	en_US
dc.subject	Target Detection	en_US
dc.thesis.degree	Master of Science	en_US
dc.title	Deep Learning for Weak Target Detection in Range-Doppler Data	en_US
ou.group	Gallogly College of Engineering::School of Computer Science	en_US