Monitoring deforestation using acoustic signals

Abstract

This paper provides a statistical method for detecting tree cutting activity using acoustic signals produced by saw scratching through a bole. An experimental setup is proposed for recording the data in real time environment. Data was collected and then processed by an SNR based algorithm to separate the noise from acoustic signals. A modified MFCC is then used to draw out the features of each five-second sample received after preprocessing. Linde-Buzo-Gray algorithm is used to fetch the statistical properties of the identified feature array. Finally, the acoustic signals are classified using Dynamic Time Warping (DTW) algorithm and half of the identified feature array and performance of the algorithm were tested by using rest of the marked feature arrays.