BIRDS-BD: A Comprehensive Dataset for Bangladeshi Local Birds and Classification Using Vision Transformer with Transfer Learning

In this research, we present an extensive exploration of machine learning techniques applied to the detection and classification of 16 distinct bird species indigenous to Bangladesh. Leveraging the advanced Vision Transformer model, our study addresses the challenge of fine-grained bird identification, encompassing species such as Bashpati, Duel, Kak, Bok, Eagle, Kobutor, Bulbuli, Chorui, Finge, Kukil, Ghughu, Masranga, Myna, Shalik, Tia, and Tuntuni with 12800 image without including augmented image. This vast size of dataset enrich the research on this topic. Our approach involves training the Vision Transformer model on a diverse dataset of bird images collected across various regions of Bangladesh. The model’s intrinsic architecture enables it to efficiently discern intricate features and patterns within the images, facilitating precise classification of the targeted bird species. We rigorously evaluate the model’s performance in terms of precision, recall, and F1-score, revealing an impressive accuracy 92.03%. This research underscores the practicality and efficacy of machine learning techniques, particularly through the Vision Transformer model, for Bangladeshi bird detection and classification. The results not only contribute to the understanding of local bird diversity but also exemplify how cutting-edge technology can play a pivotal role in wildlife conservation and research within the region.