Target shape identification based on unsupervised learning

We introduce an unsupervised-learning-based technique to analyze the Radio-Frequency (RF) scattering signals from a target and identify the shape of the target. A collection of scattering points on the target was obtained from the interceptor's RF seeker; then, the Expectation-Maximization (EM) algorithm was applied to classify them and find the statistical characteristics of each cluster. The computation results provide good estimates on the general shape of the target, which includes the length and the location of specific spots, even without sufficient prior knowledge. The proposed technique was verified via Monte-Carlo simulations using the target Radar Cross-Section (RCS) models and the probabilistic models of the seekers. The algorithms were also implemented on the embedded flight computer, and the real-time location performance under realistic environmental conditions were validated via a series of experimental tests.