Efficient Automatic Pollen Recognition From Fossil Pollen Samples: A High-Resolution Example Record From Palaeolake...

Area of Science:

• Paleoecology
• Palynology
• Computational Biology

Background:

• Pollen analysis is vital for reconstructing past environments but traditionally manual and time-intensive.
• Existing automated methods using neural networks struggle with real-world fossil pollen data from sediments.
• High-resolution, large-scale pollen analysis has been limited by manual processing constraints.

Purpose of the Study:

• To introduce and evaluate the TOFSI (Time-of-Flight Secondary Ionization) approach for automated pollen detection and classification.
• To apply TOFSI to a high-resolution lake sediment sequence for detailed paleoecological reconstruction.
• To assess the accuracy and efficiency of TOFSI compared to traditional manual methods.

Main Methods:

• Development of a novel approach (TOFSI) employing two neural networks for object detection and subsequent classification.
• Training a model to recognize 48 distinct pollen, spore, and Non-Pollen Palynomorph (NPP) classes.
• Application of the TOFSI approach to a 1 cm resolution lake sediment core.

Main Results:

• TOFSI achieved high performance (≥0.9 recall and precision at 0.5 confidence) for well-represented classes in training data.
• Performance showed a decline for classes with fewer than approximately 100 training examples.
• The method demonstrated excellent accuracy in detecting and classifying multiple pollen, spore, and NPP types in lake sediment.

Conclusions:

• TOFSI enables accurate, automated pollen analysis in lake sediments when sufficient training data is available.
• The approach offers both fully automated (limited resolution) and semi-automatic (full resolution with manual revision) workflows.
• TOFSI significantly reduces analysis time, increases count sums, enhances statistical reliability, and offers practical improvements for palynologists.