Implementation of an underwater image classifier using neural networks

It is often very difficult to classify the obscured underwater images with robust success rates using classical statistical algorithmic approaches. For such classification problems, the application of Artificial Neural Networks are found to have improved performance. This paper presents the prototype of a system for classifying underwater images into two broad categories, viz., natural shapes and anthropogenic wrecks or archaeological remnants. In the prototype system, a multilayer feed forward network, which has been trained with a large number of images to produce an acceptable level of robustness, is used for classifying the images. Different back propagation methods and a variable number of hidden layers have been attempted with the prototype neural network system for ensuring the robustness of the system.