Machine learning applied to recognition of dinoflagellate cysts: Type study with the species Batioladinium longicornutum

This study explores the application of YOLOv10, a cutting-edge object detection framework, to automate the identification and classification of Batioladinium longicornutum . Utilizing a dataset of 137 annotated images, we trained and validated the model to distinguish B. longicornutum from other species with a mean Average Precision (mAP@0.5) of 62.0 %. The methodology incorporated robust data augmentation techniques and evaluation metrics, including precision-recall analysis, confusion matrices, and cross-validation. YOLOv10's architecture facilitated accurate feature extraction and efficient classification, even with a relatively small dataset. While this study focuses on species-level identification, future work will extend to morphological and preservation state classifications, offering broader applications in automated palynology. These findings demonstrate the potential of YOLOv10 to revolutionize taxonomic workflows and enhance the efficiency of paleontological research. • Deep learning detects and classifies Batioladinium longicornutum cysts automatically. • YOLOv10 automates cyst classification, boosting efficiency in palynological research. • YOLOv10 aids preservation state analysis and feature study, enriching palynology research.