Abstract

Subbottom acoustic profiler provides acoustic imaging of the subbottom structure of the upper sediment layers of the seabed and it is commonly used in geological and offshore geo-engineering applications. Delineation of the subbottom structure from a noisy acoustic data and classification of the sediment strata is challenging with the conventional signal processing techniques. Image processing techniques utilise the spatial variability of the image characteristics, known for their potential in medical imaging and pattern recognition applications. In the present study, they is found to be good in demarcating the boundaries of the sediment layers associated with weak acoustic reflectivity, hidden in the noisy background. The study deals with application of image processing techniques, like segmentation in identification of subbottom features, extraction of textural feature vectors using grey level co-occurrence matrix statistics for different sediment strata inferred. Utilising the feature vectors, an SOM unsupervised neural network model was used in classification of the sediment type of different strata of a four layered structure inferred from an image. The model was also tested for its consistency, with repeated runs with different configuration. Successful classification of a few untrained test images representing similar environment using the network model as expected.