An application of the spectral kurtosis for early detection of subterranean termites

This paper deals with termite detection in a noisy environment using higher-order statistics (HOS). The results could be extrapolated to all impulse-like insect emissions. Sliding higher-order cumulants offer distinctive time instances, as a complement to the sliding variance, which only reveals power excesses in the signal; even for low-amplitude impulses. The spectral kurtosis reveals non-Gaussian characteristics (the peakedness of the probability density function) associated to these non-stationary measurements, specially in the near ultrasound frequency band. Contrasted estimators have been used to compute the higher-order statistics.