Texture classification of side-scan sonar images with neural networks

Presents a texture classifier for side-scan sonar image classification using two cascaded trained multilayer feedforward neural networks (acting as a principal feature extraction network and a pattern classification network, respectively). The structure of this classifier is described and a synthesised training system for constructing both networks is given. Typical experimental results are provided, showing that the incorrect classification rate of the resulting classifier is rather low. A practical application system in classifying side-scan sonar images is also presented. These experimental results, together with the inherent parallel computation mechanisms of artificial neural networks (ANNs), clearly demonstrate the applicability of the cascaded neural networks based classification technique in efficiently performing texture classification of side scan sonar image