Classification of large ornithopod dinosaur footprints using Xception transfer learning.

Area of Science:

• Paleontology
• Computer Science
• Machine Learning

Background:

• Large ornithopod dinosaur footprints are found globally, but ichnotaxonomy faces challenges due to oversplitting.
• Distinguishing valid ichnotaxa (trace fossils) is crucial for understanding dinosaur behavior and evolution.

Purpose of the Study:

• To develop an automated method for classifying large ornithopod dinosaur tracks using machine learning.
• To address and mitigate the historical problem of oversplitting in ichnotaxonomy.

Main Methods:

• Utilized a convolutional neural network (CNN) with Xception transfer learning for track classification.
• Trained the model on 274 image data of dinosaur footprints over 162 epochs.
• Evaluated model performance on an independent set of footprint illustrations.

Main Results:

• Achieved a trained model accuracy of 96.36% and a validation accuracy of 92.59%.
• Demonstrated the model's capability to correctly classify large ornithopod dinosaur footprint illustrations.
• Performance is influenced by the quality of the input data, with better results for well-preserved footprints.

Conclusions:

• The developed machine learning model effectively classifies large ornithopod dinosaur footprints, aiding ichnotaxonomic assignments.
• This approach offers a novel, academic-level application of AI to help resolve the oversplitting problem in dinosaur trace fossils.
• Future applications may benefit from diverse data types (photos, 3D scans) and advanced ML models capable of handling new classes.