YOLOv8-Based Transfer Learning for Mangrove Species Classification Using Leaf Images

Abstract

Reliable classification of mangrove species is essential for conservation efforts, ecological research, and effective biodiversity monitoring. Traditional identification methods, which depend on human observation, are resource-intensive and necessitate domain-specific knowledge. This study employs deep learning, specifically YOLOv8 with transfer learning, for mangrove species classification based on leaf images. A dataset of 4,311 leaf images, originally gathered, representing 10 mangrove species (Avicennia marina, Avicennia rumphiana, Bruguiera gymnorrhiza, Ceriops decandra, Ceriops tagal, Excoecaria agallocha, Rhizophora apiculata, Rhizophora mucronata, Rhizophora stylosa, and Xylocarpus granatum), was utilized for training and evaluating the model. The model attained a strong top-1 training accuracy of 98.3% and a testing accuracy of 97.42%. A confusion matrix analysis revealed strong classification performance, particularly for Rhizophora mucronata, Avicennia rumphiana, and Rhizophora stylosa, though some misclassifications occurred among morphologically similar species. The results highlight the potential of YOLOv8 in plant species identification, providing a scalable and efficient solution for large-scale biodiversity assessments and conservation efforts.