Automated graptolite identification at high taxonomic resolution using residual networks

Area of Science:

• Paleontology
• Artificial Intelligence
• Machine Learning

Background:

• Graptolites are vital fossils for evolutionary studies and shale gas exploration.
• Traditional graptolite identification relies on visual assessment by taxonomists, a process challenged by complex morphology and preservation issues.
• Artificial intelligence (AI) offers potential solutions for automating and improving paleontological classification tasks.

Purpose of the Study:

• To develop and evaluate a deep learning model for accurate graptolite identification.
• To compare the AI model's performance against human taxonomists in classifying graptolite genus and species.
• To assess the potential of AI in enhancing both paleontological research and shale gas exploration.

Main Methods:

• Construction of the largest professional single organism image dataset (>34,000 images, 113 species) with pixel-level annotations.
• Training and evaluation of deep learning networks for graptolite classification.
• Benchmarking AI model performance against expert taxonomist accuracy and efficiency.

Main Results:

• The deep learning model achieved 86% accuracy in identifying graptolite genus and 81% in species identification.
• The AI-based method demonstrated superior performance compared to taxonomists in accuracy, speed, and generalization.
• The model effectively recognized minute morphological details, exceeding human capabilities.

Conclusions:

• AI-powered graptolite identification can achieve taxonomist-level accuracy and beyond.
• This AI method can be integrated into accessible applications, democratizing graptolite identification.
• The technology promises to significantly enhance the efficiency of shale gas exploration and paleontological research.