Evaluating the Efficacy of Probabilistic Neural Networks to Determine Stock Structure in Sockeye Salmon Using Fourier Transformed Luminance Profiles of Scale Circuli

摘要

Patterns of circuli groupings within scales are used to determine the stock structure of sockeye salmon (Oncorhynchus nerka), The methodology typically employed involves using trained scale readers to interpret and manually measure circuli spacing patterns. These measurements are used as input into Linear Discriminant function Analysis (LDA) to determine stock structure. This pilot study introduces a new technique, probabilistic neural networks, to evaluate scale patterns for stock composition. We compare the method directly to LDA by using the same measurement data as input. We then explore Fourier analysis of luminance profiles of the scale images as an objective means to classify scale patterns. The samples used in the pilot study are from two Canadian stocks and one Alaskan stock encountered in South-east Alaskan fisheries. Correctly identifying these stocks has been a challenging problem for fisheries management.