Performance Comparison of Automated Induction-Based Algorithms for Landmine Detection in a Blind Field Test

Wideband Electromagnetic Induction (EMI) data provides an opportunity to apply robust statistical signal processing techniques to potentially mitigate false alarm rates in real-time landmine detection. This paper explores the application of matched subspace detectors (MSDs) and Support Vector Machines (SVMs) to this problem. A library of landmine responses is generated from a set of calibration data and a bank of matched subspace detectors, each designed to detect a specific mine type, is developed. Support vector machines are also considered for target/clutter discrimination. These are developed based on landmine signatures, decay rate estimates, and the outputs of matched subspace filter banks. Synthetic data sets are generated and matched subspace detectors and support vector machines are trained using this synthetic data. Receiver Operating Characteristics (ROCs) for matched subspace detectors and support vector machines are presented for both experimental and simulated data sets. The results indicate that substantial reductions in false alarm rates can be achieved using these techniques, but that simulated data sets may not provide accurate predictors of performance.