A NEURAL-NETWORK APPROACH TO CLASSIFICATION OF SIDESCAN SONAR IMAGERYFROM A MIDOCEAN RIDGE AREA

A neural-network approach to classification of sidescan-sonar imagery is tested on data from three distinct geoacoustic provinces of a midoc ean-ridge spreading center: axial valley, ridge flank, and sediment po nd. The extraction of representative features from the sidescan imager y is analyzed, and the performance of several commonly used texture me asures are compared in terms of classification accuracy using a backpr opagation neural network. A suite of experiments compares the effectiv eness of different feature vectors, the selection of training patterns , the configuration of the neural network, and two widely used statist ical methods: Fisher-pairwise classifier and nearest-mean algorithm wi th Mahalanobis distance measure. The feature vectors compared here com prise spectral estimates, gray-level run length, spatial gray-level de pendence matrix, and gray-level differences. The overall accurate clas sification rates using the best feature set for the three seafloor typ es are: sediment ponds, 85.9%; ridge flanks, 91.2%; and valleys, 80.1% . While most current approaches are statistical, the significant findi ng in this study is that high performance for seafloor classification in terms of accuracy and computation can be achieved using a neural ne twork with the proper combination of texture features. These are preli minary results of our program toward the automated segmentation and cl assification of undersea terrain.