Quantifying larval infestation with an acoustical sensor array and cluster analysis of cross-correlation outputs

An automated, computer-based system was designed to quantify, infestation of internally feeding larvae in a grain sample by obtaining data correlated with the location of sound sources, Information related to the relative arrival times of insect feeding sounds to an army of acoustic sensors is obtained despite the low signal to noise unties and the differential distortion induced by sound propagation through the non-uniform grain medium to the different sensors. This is achieved by employing parallel acquisition of all sensor outputs and cross-correlation analyses of all adjacent sensor pairs in the vicinity of the sensor with the largest signal. The peak location times of the resulting cross-correlograms cluster together for multiple sounds produced by the same insect but otherwise ave more broadly distributed. A cluster analysis algorithm was developed to group sounds with similar 'fingerprints' (i.e. patterns of peak locations across several cross-correlograms). Each sufficiently large group of matching sounds indicates the presence of an insect. Published by Elsevier Science Ltd.