AI-based coral species discrimination: A case study of the Siderastrea Atlantic Complex

Species delimitation in hard corals remains controversial even after 250+ years of taxonomy. Confusing taxonomy in Scleractinia is not the result of sloppy work: clear boundaries are hard to draw because most diagnostic characters are quantitative and subjected to considerable morphological plasticity. In this study, we argue that taxonomists may actually be able to visually discriminate among morphospecies, but fail to translate their visual perception into accurate species descriptions. In this article, we introduce automated quantification of morphological traits using computer vision (Completed Local Binary Patterns-CLBP) and test its efficiency on the problematic genus Siderastrea. An artificial neural network employing fuzzy logic (Θ-FAM), intrinsically formulated to deal with soft and subtle decision boundaries, was used to factor a priori species identification uncertainty into the supervised classification procedure. Machine learning statistics demonstrate that automated species identification using CLBP and Θ-FAM outperformed the combination of traditional morphometric characters and Θ-FAM, and was also superior to CLBP+LDA (Linear Discriminant Analysis). These results suggest that human discrimination ability can be emulated by the association of computer vision and artificial intelligence, a potentially valuable tool to overcome taxonomic impediment to end users working on hard corals.