Ellipsoidal Basis Functions for Computer Aided Identification of Mines

The objective of this proposal is to explore and develop a candidate identification algorithm, the Ellipsoidal Basis Functions method and demonstrate capability for computer-aided mine identification with specific exploitation of 3-D information.Currently, mine identification systems rely on post mission processing to complete mining hunting operations. Real-time computer aided detection and computer aided classification (CAD/CAC) is just now being pursued by many researchers in order to reduceoperator workloads. We propose to develop the EBF method, utilizing AQS20(A) and other acoustic and or magnetic sensor data to provide an enhanced target discrimination capability as compared to the standard Bayesian Estimation, Correlation/Matched-Filter,or Energy Classifier methods. We note that a significant feature of the EBF method is that it seamlessly lends itself to fusing data from multiple sensors as discussed below and that the EBF method is a fully data-driven method. The method does not requireany physical or environmental models or parameters of the observer system. These can be incorporated in the pre-processor if a prior information is available. Also, the EBF method can readily exploit 3-D data since the data is represented as featurevectors, the dimensionality of the data can be arbitrary. The proposed investigation will provide valuable tools and validation for applications of the Ellipsoidal Basis Functions method to areas other than mine-hunting also such as missile and homelanddefense as well as object and face recognition technologies. There currently exist various face recognition technologies that have made way into some applications (e.g. Connecticut DMV, West Virginia Child ID programs). The deployed software is still beingtested and refined. Most of the techniques currently use a principal components analysis (PCA) followed by extraction of