Invariant Morphological Descriptors from Otolith Shape in Environment Automatic Classification

One of the ways to recognize patterns is through the intrinsic information of the object's shape by descriptors that allow us to quantitatively describe the contained shape of the object. In ichthyology, the otolith shape recognition is notable for each fish species, which allows the study of the sagitta otolith in species classification, the comparison of otolith shape across the fish ontogeny or growth, and the symmetry analysis of these structures in the case of the same fish. In the last twenty years, there has been a valuable contribution regarding otolith shape analysis and various types of morphometric descriptors have been proposed. The first objective of this work is to propose the implementation of invariant morphometric descriptors, as Discrete Compactness, Discrete Tortuosity, Non-Circularity, and Mirror-Symmetry, and compare their performance with other reported morphometric descriptors. The second objective is the implementation of the Random Forest (RF) algorithm to classify the fish species according their environment (marine, brackish, and freshwater). The right and left sagittae otoliths of 139 marine, brackish, and freshwater species (adults and juveniles) of the Yucatan peninsula and Gulf of Mexico were analysed with invariant and other reported descriptors. The global results show that the invariant descriptors can provide complementary information to other reported descriptors based on area or perimeter, given a low correlation between these features. The environment classification of species using a RF classifier showed that 83% of species correspond positively with their environment classification. This classification analysis can be a useful tool for studies of trophic dynamics, or in archaeological and paleontological studies on fossil fauna this classification tool would allow inferring from remains the environment of the studied communities and their evolution over long periods of time.