Wildlife intruder detection using sounds captured by acoustic sensors

In this paper we classify the sounds originated from humans, birds and cars. The motivation of such a classification is to detect the intruders into protected wildlife regions such as protected forests, lakes, and other natural reservations. The proposed algorithm for sound encoding and classification is Time Encoded Signal Processing and Recognition (TESPAR) combined with the archetypes technique. We have tested our method on a database consisting of 300 recordings, 100 for each class, and several types of noise (white Gaussian noise, rain sound and wind sound) have been added to the recordings in order to simulate the different outdoor environments. Several pre-processing steps have been included and tested in order to verify the improvement of the classification accuracy. We performed a downsampling from 8 kHz to 6 kHz of the original recordings, followed by band pass filtering and the results shown an increased efficiency of TESPAR in the classification process.