WingSpan: Advanced Butterfly Classification using ResNet-18 Networks

Butterfly classification is a challenging task due to the intricate patterns and diverse morphological characteristics exhibited by various species. This study leverages the ResNet-18 DL model to address the complexities of butterfly classification, aiming to achieve high accuracy and robust performance across diverse datasets. The ResNet-18 model, known for its residual learning framework, is particularly effective in training deeper networks by mitigating the vanishing gradient problem. A dataset comprising high resolution images of 75 butterfly species was used, with images split into training, validation, and test sets. The model was trained over 40 epochs, incorporating data augmentation and optimization techniques to enhance generalization. The results demonstrate that the ResNet-18 model achieved a training and validation accuracy approximately 98% , with minimal overfitting as indicated by the convergence of the training and validation loss curves. The findings underscore the efficacy of the ResNet-18 model in accurately classifying butterfly species, providing a foundation for future research in automated species identification and contributing to ecological monitoring and biodiversity conservation efforts. The study illustrates the potential of DL models in addressing complex classification tasks in entomology and beyond.