Feature and decision level sensor fusion of electromagnetic induction and ground penetrating radar sensors for landmine detection with hand-held units R.J. Stanley a, * , P.D. Gader b , K.C. Ho c a Department of Electrical and Computer Engineering, University of Missouri, Rolla, MO 65401, USA b Department of Computer Engineering and Computer Science, University of Missouri, Columbia, MO 65211, USA c Department of Electrical Engineering, University of Missouri, Columbia, MO 65211, USA Received 14 July 2001; received in revised form 19 November 2001; accepted 25 February 2002 Abstract Strategies for fusion of electromagnetic induction (metal detector (MD)) and ground penetrating radar (GPR) sensors for landmine detection are investigated. Feature and decision level algorithms are devised and compared. Features are extracted from the MD signals by correlating with weighted density distribution functions. A multi-frequency band linear prediction method generates features for the GPR. Feature level fusion combines MD and GPR features in a single neural network. Decision level fusion is performed by using the MD features as inputs to one neural network and the GPR features as inputs to the geometric mean and combining the output values. Experimental results are reported on a very large real data set containing 2315 mine encounters of different size, shape, content and metal composition that are measured under different soil conditions at three distinct geographical locations. Ó 2002 Elsevier Science B.V. All rights reserved. Keywords: Hand-held unit; Landmine detection; Signal processing; Image processing; Pattern recognition; Sensor fusion; Electromagnetic induction; Ground penetrating radar 1. Introduction The United Nations estimates that tens of millions of mines lie buried around the world. The detection of landmines is an important mission to save lives for many innocent victims. Many currently fielded landmine de- tection systems consist of hand-held metal detectors (MDs). MDs have proven utility in detecting metal- cased landmines. However, many modern landmines are plastic-cased and have small amounts of metal. They can produce weak responses in MDs that are difficult to distinguish from responses from metal clutter. Ground penetrating radar (GPR) sensors can detect plastic-cased mines. It is therefore desirable to investigate their utility for landmine detection. GPR sensor systems typically suffer from high false alarm rates since they respond to dielectric discontinuities in metallic and non-metallic objects. In this research landmine detection will be ex- plored based on hand-held units containing single coil MD and a stepped frequency, continuous wave GPR. Weak responses from a MD may be useful for re- jecting GPR false alarms due to non-metal dielectric discontinuities while retaining GPR detections of low- metal mines. Therefore, fusion of MD and GPR offers the potential for achieving higher probabilities of de- tection and lower false alarm rates that can be achieved with either sensor alone. In this paper, feature and de- cision level fusion algorithms for fusion of MD and GPR for landmine detection using data acquired from a hand-held system are devised and compared. Bayesian and other statistical analysis techniques have been applied to time- and frequency-domain EMI sensor data [1–5]. The basis for landmine detection was modeling target response based on EMI representations including signal energy and exponential decay rates. Milisavljevic et al. have investigated landmine shape and model specific factors relating to sensor utilization and fusion for landmine detection [6–8]. There are several * Corresponding author. Address: Department of Electrical and Computer Engineering, University of Missouri, 1870 Miner Circle, 127 Emerson Electric Company Hall, Rolla, MO 65409-0040, USA. Tel.: +1-573-341-6896; fax: +1-573-341-4532. E-mail address: stanleyr@umr.edu (R.J. Stanley). 1566-2535/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved. PII:S1566-2535(02)00071-4 Information Fusion 3 (2002) 215–223 www.elsevier.com/locate/inffus