Energy, Vibration And Sound Research Group (e-VIBS) School Of Science And Technology Universiti Malaysia Sabah : Bioacoustics Signal Modeling Using Time-Frequency Distribution

• Main Authors: Jedol Dayou, Ng, Chee Han, Ho, Chong Mun, Abdul Hamid Ahmad, Mohd Noh Dalimin, Sithi V. Muniandy
• Format: Research Report
• Language: English; English
• Published: 2011
• Subjects: QA76.75-76.765 Computer software; TA365-367 Acoustics in engineering. Acoustical engineering
• Online Access: https://eprints.ums.edu.my/id/eprint/30823/1/Energy%2C%20Vibration%20And%20Sound%20Research%20Group24%20PAGES.pdf; https://eprints.ums.edu.my/id/eprint/30823/2/Energy%2C%20Vibration%20And%20Sound%20Research.pdf; https://eprints.ums.edu.my/id/eprint/30823/
• Institution: Universiti Malaysia Sabah

Biodiversity is one of the major studies in bio-conservation, which enable to evaluate the quality of ecosystem in a specific area, especially for protected area. In order to monitor the quality of the ecosystem structure, a long term rapid diversity assessment is needed. In term of that, bioacoustics has been introduced as a beneficial method for local species richness estimation. However, this method is still in the infancy state and many improvements are needed for more practical purposes. This research is carried out to develop new bioacoustics species identification method with the improvement in the identification accuracy. The method which developed in this research is based on entropy principles, and implements on Fourier transform (FT) and wavelet transform (WT) of bioacoustics signal. Several entropy principles including Shannon, Renyi and Tsallis, are investigated which representing measurement of richness of the information contents and complexity of a bioacoustics signal. To evaluate the new identification system, nine frog species from Microhylidae family was selected as test samples. Ten syllables were segmented from each frog sounds and characteristic of each syllables was extracted with the corresponding features which carried out in this research. All of the test samples were then sent into the k-nearest neighbor (k-NN) classifier for classification purpose. The k-­NN classifier compared the test samples with the training data set in order to recognize and identify the frog species. To establish a base trial data, the widely used spectral centroid (SC) and wavelet centroid (WC) were used as reference. The SC and WC of the syllables for each species were determined. It is found that, in terms of the average of classification accuracy for all test samples, the SC method has shown slightly better compared to the WC method. The classification results in average were 88.89% for the SC features and 86.67% for the WC features. The entropy alone, if implemented on raw bioacoustics signal shows reduction rather than improvement in the identification accuracy compared to the reference (SC and WC). It is found for example that the average identification accuracy for Shannon entropy (SE), Renyi entropy (RE) and Tsallis entropy (TE) were 76.67%, 75.56% and 83.33%, respectively. Due to the poor classification results of the entropy alone approaches, alternative methods were proposed in this work called wavelet entropy (WE). WE is a combination of interdisciplinary concepts between wavelet transform and entropy. In order to archive this, the entropies (SE, RE and TE) were extracted from three types of wavelet transforms, namely continuous wavelet transform (CWT), discrete wavelet transform (DWT) and wavelet packet decomposition (WPD), of a bioacoustics. signal. In this work, two possible ways to extract the entropy from CWT were introduced, which were wavelet scale entropy (WSE) and wavelet time entropy (WTE). Entropy that extracted from DWT and WPD were called as discrete wavelet entropy (DWE) and wavelet packet entropy (WPE), respectively. The species identification results based on these WE features which extracted from the bioacoustics signals were then examined and compared. In term of SE approach, WPE has given the best classification result compared with others (WSE, WTE and DWE), which was over 98% of accuracy. However, WSE was the best method for the RE approach with the accuracy of 92%. Based on TE approach, WPE has shown the best result with the classification accuracy of 100%. In term of that, this research work has proven that the WPE is the best method in the TE approach for species identification on bioacoustics signals. In conclusion, this work has successfully developed the species identification system based on bioacoustics signals by using the concept of WE. By comparing to the reference methods (conventional or classical methods), this study has proven that the performance of the bioacoustics species identification system can be improved by using the entropy approach with association of WT. Since the bioacoustics species identification system that proposed in this study is based on entropy approach, the computer algorithms is much easier (less complex) compared to the conventional methods, particularly based on spectrogram and sonogram. The proposed method can reduce the energy and time consumptions in terms of data processing.