Utility of Computer Assisted Age and Growth Estimation (CAAGE) in the analysis of otolith images for separating herring (Clupea harengus) spawning groups.

Building on recent work reporting successes of fully automatic CAAGE systems applied in aging of commercially important demersal fish species this study investigates the feasibility of using the technology for stock discrimination. We review the typical imaging pipeline of CAAGE systems highlighting popular algorithms and potential pitfalls. We review morphological image processing techniques for identifying the otolith core and we compare signal processing algorithms for analysing 1-D transept profiles originating from the core and extending to its edge. We discuss practical problems that arise when applying these approaches to species of pelagic fish and show that the methods are sensitive to parameter selection and other factors associated with image acquisition. Then we discuss how systems that employ an interactive user interface overcome some of the problems encountered with fully automated systems, while still providing benefits in terms of improved efficiency. To illustrate this work we focus on a benchmark problem concerning the reassignment of Thames and North Sea herring individuals to their source populations by measurement of year 1 incremental growth. We perform interspecies classification experiments using linear discriminant analysis (LDA) on measurements made on otolith images by an expert reader, an automatic system, and an interactive system. Our fully automatic system achieves an accuracy of >70% while an expert working with conventional image annotation tools can achieve >85%. Our results demonstrate that an expert working interactively with a computer assisted age and growth estimation system can quickly correct errors made by the system and improve efficiency.