Automatic Frog Calls Monitoring a Machine Learning Approach
Abstract
Automatic recognition of frog vocalization is considered a valuable tool for a variety of biological research and environmental monitoring applications. This thesis proposes to develop an automatic, unattended monitoring system which can recognize the vocalizations of four species of frogs in the State of Oklahoma and can identify different individuals within the species of interest. The proposed monitoring system deployed one directional microphone to record the frog calls in the field continuously. Sound signals were stored in digital audiotape first and then transmitted into a PC with WAVE file fonnat. Species identification was perfonned first with the proposed filtering and grouping algorithm. Individual identification, which can detect different individual frogs within the same species, was perfonned in the second stage. Digital signal pre-processing, feature extraction, feature vector dimension reduction and pattern classification were perfonned step by step in this stage. Different feature extraction algorithms, induding the time domain method (Linear Predictive Coding), the frequency domain method (Time-Dependent Fourier transform), and time-scale domain method (Wavelet Packet Transfonn), and two different dimension reduction algorithms are synergistically integrated to produce final feature vectors which were to be fed into a neural network classifier. The simulation results show the promising future of deploying an array of continuous, on-line environmental monitoring systems based upon nonintrusive analysis of animal calls.
Collections
- OSU Theses [15752]