Abstract:

Fast discrete curvelet transform is performed on the seabed sonar image to obtain low frequency subband coefficients and various directional subband coefficients. Standard deviation is used to describe the image’s global unevenness in the low frequency subband. The texture energy measure (TEM) is used to process coefficients at each scale in the directional subbands, which describes directivity and roughness of texture. The extracted texture feature vectors are used in the classification of side-scan sonar seafloor images with support vector machines (SVM). The side-scan sonar images of sand, mud and rock seafloors are classified using the method described in this paper, and other methods. Comparison results show that the presented seafloor classification method has better performance.

Key words: side-scan sonar imagery, fast discrete curvelet transform (FDCT), texture feature, SVM, sediments classification