MSFF: Multi-Spectral Feature Fusion

We propose to build a Multi-Spectral Feature Fusion Framework (MSFF): a framework for fusion of data from Multi-Spectral sensors. The ultimate goal of the proposed solution is to improve the probability of target detection and identification. Two major innovations of the proposed effort are: (i) Composite feature vector that includes diverse features such as shape parameters (area, boundary length, number of edges, etc.), multi-scale texture parameters, moments of the radiance intensity of different orders, radiance energy of the image determined in multi-scale wavelet coefficient domain, and (ii) Innovative approach to closed loop association and classification that will apply machine learning methods to establish "similarity" in the detected features of the object and their "proximity" to a reference entity. The MSFF framework will support the comprehensive lifecycle from data preprocessing to transformation, feature extraction, association, and identification. The Phase I project will develop the concept of operation and algorithms. The algorithms will be tested and analyzed on simulated data using a testbed of physical and engineering models of sensors. The Phase II project will develop and harden software for near-real-time data processing.