Automatic acoustic mine detection using morphological perceptions

Recent developments in landmine detection based on the acoustic-to-seismic coupling phenomenon have demonstrated the feasibility to detect both metallic and nonmetallic mines. In this method, a loudspeaker above the ground surface insonifies the target region of the surface. Acoustic energy is coupled into the ground producing ground vibrations. The ground vibration velocity at the ground surface is measured with a laser Doppler vibrometer producing a ground surface image whose pixel values are the measured velocity amplitudes. Vertical particle velocity amplitudes directly above a mine contrast with those of the background (‘‘away from the mine’’). Image processing techniques are used in order to automatically detect regions of interest (‘‘possible mines’’). Further image analysis methods extract geometric as well as amplitude features in order to produce four-dimensional feature vectors. These vectors are input to a novel neural network based on mathematical morphology which classifies the regions of interest into mines and false alarms. After training the network, the network correctly identified all the mines on a given test set with an extremely low alarm rate.