A portable terminal for acoustic monitoring and online recognition of bats with CNN-LSTM

Abstract

The acquisition and recognition of ultrasonic signals serves as pivotal mechanisms for the dynamic monitoring of bat species. In this study, we introduce a novel portable terminal for ultrasonic monitoring and online recognition of bats, leveraging an embedded platform in conjunction with the AudioMoth device. This research capitalizes on the distinctive differences observed in the echolocation signals’ typical characteristics across various bat species, alongside their spectrogram features. To this end, a sophisticated voiceprint recognition method was developed, combining the strengths of convolutional neural network with long short-term memory network. This method was subsequently integrated into the portable terminal. Furthermore, the Majority Vote Algorithm was employed to improve the recognition accuracy. Experimental results obtained from trials conducted within a controlled bat laboratory environment demonstrate the terminal’s capability for real-time collection and online recognition of bat ultrasonic signals. Remarkably, the system achieved a recognition accuracy of 99.18%, surpassing the performance metrics of four conventional deep learning models typically employed in similar contexts. This research not only provides a practical case for the acoustic monitoring and recognition of bat species but also holds the potential for broader application in wildlife diversity investigations.