Machine Learning for Robust Automatic Target Recognition

Toyon Research Corporation and Dr. David Miller from The Pennsylvania State University propose to develop a robust automatic target recognition (ATR) module that is capable of classifying ground vehicles observed by radar and/or multispectral sensors. The classifier that will be developed will also have the ability to be applied to any general features extracted from the signal of any sensor. We will develop the ATR using a semisupervised learning algorithm that is an extension of the expectation maximization (EM) technique. We demonstrated the feasibility of this approach in Phase I using signatures of ground vehicles observed by a radar operating in an HRRGMTI mode. We demonstrated the ability to recognize known targets as well as the ability to recognize that a new observation did not belong to any of the target classes for which the ATR was trained. We will continue this work in Phase II by improving the statistical modeling process and the model selection process so that the ATR is better able to discover new classes of targets. Additionally, we will develop an ability to fuse multiple measurements from different sensor types to arrive at an effective classification based on all of the available data. We will also integrate our ATR algorithms with an existing Toyon tracker to demonstrate the feature-aided tracking potential of this new ATR algorithm. We will demonstrate the performance of our algorithms through a variety of experiments using radar and multispectral data.