碩士 / 國立臺灣海洋大學 / 電機工程學系 / 103 / With the growth of the advanced methods and technology, scientists keep nailing the enigmas of ocean environment using a variety of methods and techniques. The most familiar method in the ocean research about the sound analysis is using the time-frequency analysis to reflect the features of sound messages from the corresponding frequency spectra. However, at the same time of sound recording, a number of factors, such as raindrops, winds, other living beings in nature and artificial ship noise, have adverse effects on the analysis of the target’s sounds, so it would be hard to identify sound features by only performing time-frequency analysis.
In this paper, we focus on two audible sounds, Click and Whistle, of bottlenose dolphins for analysis. In order to avoid background noise from interfering with sound feature analysis, a preprocessing step of background removal is performed and the frequency of the largest amplitude spectral value is extracted as the first feature from
the resultant spectrogram. Next, the binary segmentation method is executed to obtain the percentage of the white pixels as the second feature, and then the image processing methods of infill, opening operation, and edge detection are applied to the spectrogram to obtain the image feature contour, from which the first five largest time intervals are found as another features via vertical projection, and finally, the k-mean method is employed for effective classification. In the experiment, we adopt three samples of “Click” sound and four samples of “Whistle” sound for classification, from which seven features are extracted per sample. The result of the k-mean method shows that the correct classification rate is 71.43 % .
Identifer | oai:union.ndltd.org:TW/103NTOU5442042 |
Date | January 2015 |
Creators | Su, Po-Yu, 蘇柏宇 |
Contributors | Hung, Hsien-Sen, 洪賢昇 |
Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
Language | zh-TW |
Detected Language | English |
Type | 學位論文 ; thesis |
Format | 52 |
Page generated in 0.0072 seconds