Intelligent optical otolith classification for species recognition of osteichthytes.

Otolith study is a well-established source of information for understanding the life of fish and fish populations. Conducting fish species identification from otolith samples found in the stomach contents of marine animals finds interesting applications such as dietary studies, stock monitoring and management. Fish species identification can produce useful data for climatology and archaeology, as otoliths can be sourced from geological sediments or archaeological excavations. An automated otolith classification can assist a wide variety of scientific research. This paper presents the development of an automated fish specied identification system. The main focus of this investigation is on the commercially interesting fish of the Northern Aegean Sea. The methodology described in this paper exploits the inherent shape variability of fish otoliths according to their corresponding species. This is based on the processing and analysis of images acquired using a stereoscopic microscope fitted with a digital camera. A feature vector is then constructed that describes the morphology as well as the image statistics of the otoliths. The recognition is carried out by an intelligent classifier made of artificial neural networks. Several configurations of multi-layer perception neural networks are tested in pursuit of a practical and expandable classification system.