An acoustic Wireless Sensor Network for remote monitoring of bird calls

Abstract

Wireless Sensor Network (WSN) based acoustic monitoring is useful for ecologists for the purpose of monitoring real-time wildlife behavior across remotely located large areas, for long periods and under variable weather/climate conditions. However, stringent requirements for intense data processing and data bandwidth across network nodes makes applicability of WSN based monitoring limited. This paper presents, a mesh network of wireless sensor nodes that only requires readily available hardware, algorithms and methods for acoustic source type identification. A node was constructed by interfacing three separate modules to a microprocessor: (i) a microphone coupled to a USB sound card for recording acoustic data, (ii) a radio frequency (RF) transceiver for inter-node and remote server communication, and (iii) an external Real Time Clock module for time-synchronization between nodes. The system was designed for real-time identification of calls of Black-Rumped Flameback (BRF), a type of woodpecker endemic to the Indian sub-continent and found in parts of Sri Lanka. Any acoustic signal above a predefined intensity threshold was recorded while BRF calls were discriminated with respect to two other known bird calls. This was achieved based on a threshold estimated by measuring the cross-correlation between two known BRF calls. This method was able to successfully identify 83% of the tested BRF calls. Several more frequency domain features were also identified that are compatible with Support Vector Machines (SVM), which is a more robust identification algorithm. The SVM based was able to successfully identify 91% of the tested BRF calls. The total cost of hardware used per node was estimated to be under USD 75.