| Summary: | 碩士 === 國立臺灣大學 === 海洋研究所 === 99 === In order to objectively analyze side-scan sonar image, we use SwathView which produced by QTC in Canada to process acoustic classification. We also use SwathView to analyze the data of area between Taiwan and Penghu Island. The main feature of SwathView is segmentation method; it combines many statistic algorithms and is useful to process large dataset. Every data has its own acoustic property; we classify the data which has the same acoustic property as one class and the others are so on. Firstly, we load each file and compensate it. Secondly, we segment sonar image by small rectangle, each small rectangle area has many backscatter intensity information. Thirdly, we calculate each rectangle area’s intensity value by statistic algorithms; each rectangle area has 29 statistic values. The last, we reduce 29 values to 3 principle values which called Q1, Q2, Q3 by PCA, according to these Q values we can define each rectangle area’s acoustic property and use K-means cluster method to classify each data.
After classification by software, the result seems not well because some class has fake information. The most common fake information is along-track feature. Therefore we must process quality control. I devise the process of quality control by two steps: filter data before analysis and process data after analysis. After quality control we can get a good acoustic classification image. Then we can define each class by real habitat samples.
|
|---|
