Towards Automatic Butterfly Species Recognition Using a Single Spatio-Hyperspectral Image

Butterfly Species identification accounts nowadays for a challenge to evaluate the biodiversity state. Using special compact Hyperspectral Cameras for this task is more attractive. Whereas usual techniques use a sequence of images to compute a datacube, we focus here on a single image resulting in a partial butterfly datacube. With a pre-identification of the features from a butterfly library, we propose a combined probabilistic clustering technique based on a weighted combination of Z-score and Gaussian Naive Bayes probability, which aims to recognize the associated cluster from the particular butterfly species. Results obtained in this context achieve good performance with respect to Gaussian Naive Bayes probability or Z-score-based techniques.