Model-based statistical signal processing using electromagnetic induction data for landmine detection and classification

Traditionally, electromagnetic induction (EMI) sensors are operated in the time-domain and the response strength is related to the amount of metal present in the object. These sensors have been used almost exclusively for landmine detection. Unfortunately, there is often a significant amount of metallic clutter in the environment that also induces an EMI response. Consequently, EMI sensors employing detection algorithms based solely on metal content suffer from large false alarm rates. A second issue regarding processing of data collected on highly cluttered sites is that anomalies are often in close proximity, and the measured EMI signal consists of a weighted sum of responses from each anomaly. To mitigate the false alarm problem, statistical algorithms have been developed which exploit models of the underlying physics for mines with substantial metal content. In such models it is commonly assumed that the soil has a negligible effect on the sensor response, thus the object is modeled in "free space". To date, such advanced algorithms have not been applied specifically to the problem of detecting low-metal mines in a cluttered environment. Addressing this problem requires considering the effects of soil on signatures, separating the multiple signatures constituting the measured EMI response as well as discriminating between landmine signatures and clutter signatures. In this paper, we consider statistically based approaches to the landmine detection and classification problem for frequency-domain EMI sensors. We also develop a preliminary statistical approach based on independent components analysis (ICA) for separating the signals of multiple objects that are within the field of view of the sensor and illustrate the performance of this approach on measured data.