A Study on the Pattern Recognition of Sidescan Sonar Images using Artificial Neural Network

• Main Authors: Kun-feng Chang, 張坤鳳
• Other Authors: Yung-chung Wei
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/88014932228947761513

碩士 === 義守大學 === 資訊管理學系碩士班 === 94 === Currently, the search and detection of undersea objects mainly rely on the manual recognition of digital data obtained from optical, acoustic, or some other instruments, e.g., side scan sonar, bottom profiler, and magnetometer. For this purpose, large amounts of precious time, man-power, and equipment are consumed to recognize undersea objects. The scope of this study is to introduce and examine the application of wavelet transform and artificial neural network on side scan sonar images, and to further improve the automation and efficiency of undersea object recognition. This study utilizes the two-dimensional discrete wavelet transform and multi-resolution analysis to decompose side scan images, of which the high frequency component is capable of enhancing the pattern and edge of undersea objects. The output of multi-resolution analysis is used to increase the dimension of side scan images, hence increasing the input vector dimension of artificial neural network. To recognize and classify undersea objects, this study chooses the most popular neural network, the back-propagation network, with the steepest gradient method to minimize the errors between input and output images, and to decrease the false alarms of undersea object classification. The experiment shows that various network training coefficients, input data dimension, and wavelet transform methods do effect on the classification of artificial neural network. The back-propagation network has the better capability in pattern recognition, while compared with traditional statistic methods, i.e., the maximum likelihood classifier and the minimum distance classifier. The result of this study presents the efficiency of automatic pattern recognition, which will contribute to bathymetry studying, environmental monitoring, undersea object searching, and rescuing.