Computer vision-based transfer learning techniques for classification of local pigeon species in Bangladesh:

Abstract

In the realm of avian conservation, this thesis embarks on a pioneering journey to enhance the classification of pigeon species within Bangladesh. Leveraging the powerful Xception model, we present a breakthrough approach that attains an exceptional testing accuracy of 99.47% and minimal loss of 0.025. Our study encompasses a comprehensive dataset of 7500 images, spanning 15 pigeon species, and employs transfer learning for swift and reliable classification. While the results underscore the efficacy of our approach, the study acknowledges the challenge of subjective criteria in species classification and calls for future exploration into enhancing interpretability. Ethical considerations are central to our findings, advocating transparent communication with conservationists and the establishment of stringent ethical guidelines for responsible technology application in avian conservation. This research, a significant stride at the intersection of technology and ethics, not only contributes to avian conservation but also lays the groundwork for future investigations, paving the way for a sustainable future in avian species management and urban biodiversity preservation.