Single and Multi-Species Bird Identification Using YOLOv8 and Deep Neural Network Ensembles

Fine-grained bird species identification from images plays a critical role in biodiversity monitoring, ecological research, and wildlife conservation. However, manual identification is time-consuming, expertise-dependent, and difficult to scale across large geographic regions. This work presents a unified deep learning framework for automated single and multi-species bird identification using image data. The proposed pipeline integrates object detection and fine-grained classification by combining YOLOv8 for bird localization with multiple convolutional neural network classifiers, including ResNet-50, DenseNet-121, and MobileNetV2, for species recognition. The CUB-200-2011 dataset is standardized through YOLO-format bounding-box annotations and an ImageNet-style directory structure to support both detection and classification tasks. Detected bird regions are independently classified, and a majority-voting strategy with confidence-based tie-breaking is employed to generate the final species prediction. Experimental evaluation demonstrates that DenseNet-121 achieves the highest validation accuracy of 78.44%, followed by MobileNetV2 (76.95%) and ResNet-50 (76.07%). The hybrid ensemble approach improves robustness in complex scenarios involving multiple birds, occlusions, and visually similar species. The proposed framework provides a scalable and reproducible solution for automated fine-grained bird identification, supporting large-scale ecological monitoring, biodiversity assessment, and conservation initiatives.