Evaluating and improving a semi‐automated image analysis technique for identifying bivalve larvae

Abstract

Knowledge of the distribution, abundance, and transport of bivalve larvae is limited due to their small size, similar morphologies between species, and lack of an automated approach for identification. The objective of this research is to evaluate and improve the accuracy of ShellBi, a novel supervised image classification method that uses birefringence patterns on the shells of bivalve larvae under polarized light to identify species. The performance of the ShellBi method was tested by rearing Crassostrea virginica (eastern oyster) larvae at different temperatures (21.3 and 27.5°C) and salinities (10.3, 14.1, 14.4, and 20.5). Differences in rearing temperatures resulted in differences in classification accuracy, as did large variations in salinity (≥10 units). classification accuracies increased from 67–88% to 97–99% when training sets included images of larvae reared in conditions similar to those of the larvae being classified. Additional tests indicate that misclassification rates ranged from 0 to 13% for false positives and from 0 to 22% for false negatives, depending on the proportion of oyster larvae in the sample. Results suggest that this technique could be applied to field samples with high accuracy as long as the images that are used to make classifications include larvae that were reared in conditions that are similar to those in situ. In addition, these findings demonstrate that the ShellBi method can be used to measure and identify bivalve larvae in a different system than the one for which it was developed, suggesting that the method has broad applicability in marine and estuarine systems.