|Advanced Machine Learning ATR using SAR/ISAR data|
|Sensors & Electronics Technology|
3D CAD models, ATR, Convolutional Neural Network, Deep Learning, EM prediction codes, identification, ISAR, Machine Learning, modeling, recognition, SAR, simulation
This TG is proposed as a follow-on of SET-215 “Model Based SAR ATR”.
ATR based on high-resolution SAR (Synthetic Aperture Radar) imaging has a potential for supporting NATO operations with enhanced surveillance capabilities, including detection, classification and identification of targets. Meanwhile, progresses in the field of optical & speech recognition have made a recent breakthrough, and the radar signal processing community is starting to gather interest on these machine learning techniques, which includes (Deep) Convolutional Neural Networks (CNN) and make great use of high-end GPUs and high-performance computing clouds/nets. One of the challenges of these techniques is that they typically require a large amount of data for training. Simulated data, generated from models, may be one solution to populate these datasets.
This new TG proposes to focus its work on various aspects of modern Machine Learning and advanced signal processing for SAR/ISAR ATR, namely, but not restricted to, the use of (Deep) Convolutional Networks and the influence of their usage on recognition performances.
Comparison of real and modeled signatures of various reference targets, using multiple simulators or simulation schemes, and different CAD models or target variants.
ATR performance metrics definition. Interest of modern Machine Learning algorithms for detection/classification/identification in radar imaging.