Sediment Classification using Spectral Features of Side-Scan Sonar Images

Accurate classification of underwater sediment is important in many applications like dredging, study of marine biology, coastal engineering, and hydrography. Active Sound Navigation And Ranging (SONAR) systems are widely used in the classification of sediments. Commercially available active SONAR systems include Single-Beam Echo Sounder (SBES), Multi-Beam Echo Sounder (MBES), and Side Scan SONAR (SSS). The proposed work presents the classification of sediments using Side Scan SONAR (SSS) images using Discrete Wavelet Transform (DWT). To reduce the dimension of the feature vector we used the principal component analysis (PCA) technique which leads to low computational cost and memory requirement. The performance of Support Vector Machine (SVM) classification using three different kernel functions viz. Linear, Polynomial, and Gaussian Radial Basis (GRB) is presented in this paper. Here we have used the EdgeTech DF1000 SSS image database obtained from project REBENT, IFREMER (Location: France). The result showed that using PCA and SVM with GRB kernel function achieves 98.3 % accuracy and is well suitable for all sediment types.