Fish age analysis and classification with kernel methods

Abstract

Knowledge of age in fish populations is of great significance in stock assessment. Fishery scientists have traditionally used fish otoliths, i.e. calcified structures in the inner ear to estimate fish age, because their shape changes during a fish’s lifetime. However, many factors influence changes in otolith shape, sometimes in unpredictable ways, so manual classification remains a complex task that may yield inaccurate age estimations. Here, we discuss automatic pattern analysis and recognition for accomplishing this task and present a novel application of two well-known kernel methods: kernel principal component analysis and multi-class support vector machines for analyzing and classifying fish ages from otolith images. Significant results were achieved as a result of analyzing a cod otolith database.