Classification of fish body parts in an underwater environment

Abstract

This master thesis is the result of work conducted over the course of a semester with the goal of invstigating a possible approach to recognizing fish parts in a video stream from a camera system situated in an underwater environment. This task is seen as the first part of a three-step scheme for implementing an automatic system for fish health assessment in the fish farming industry. This thesis describes work done in setting up an interface to an IP camera that is situated in an underwater environment, collecting and labelling image material from the camera system for training and testing object classifiers, and training the object classifiers for multi-class object recognition based on image descriptors suitable for an underwater environment. Finally, a complete object recognition framework is implemented and performance tests are performed based on the pre-trained classifiers, and the results are analyzed. The results of this thesis show that it is possible to create a system that is able to perform this classification by relying on svm classifiers based on adaptations of the lbp image patch descriptor. By using a linear svm classifier good results are achieved.