Automatic Identification of Sri Lankan Butterfly Families using Ensemble Deep Learning

Butterfly diversity plays an important role in the ecosystem. Typically, experts undertake the visually demanding and error-prone task of identifying butterfly species, which requires extensive professional expertise and time. Introducing an automated algorithm becomes crucial to streamline butterfly detection, addressing this challenge and contributing significantly in conserving keystone species while advancing taxonomy and biodiversity research. This study introduces an innovative approach that employs deep learning techniques, specifically an ensemble comprising VGG16, VGG19, and ResNet50 architectures, for the automated identification of butterfly families in Sri Lanka. Despite the country's rich butterfly species diversity, previous research on automatic identification at the family level has been lacking. We compile a dataset of 2,182 images representing five Sri Lankan butterfly families (Nymphalidae, Papilionidae, Pieridae, Hesperiidae, and Lycaenidae) and use an ensemble deep learning algorithm to capture distinctive features relevant to Sri Lankan butterfly families. This is the first attempt using deep learning techniques for Sri Lankan butterfly identification. We evaluate the model performances of VGG16, VGG19, ResNet50, as well as the ensemble of these three methods. The ensemble method demonstrated an impressive accuracy of 95% for aug-mented datasets. These results highlight the effectiveness of these deep learning models in accurately identifying butterfly families within the context of Sri Lanka.