Abstract:The automation of fossil pollen analysis promises many advantages in handling large numbers of samples with less resource allocation. However, automation is often obstructed by the high abundance of organic and minerogenic non-pollen debris in fossil pollen samples. We used a Convolutional Neural Network-based approach to detect pollen-like objects in digital images of prepared microscopic slides for fossil pollen analysis and subsequently classified them into nine pollen classes and the marker spore Lycopodium . We trained the object detection and the classification model independently with a newly developed dataset of annotated images of fossil pollen grains. The object detection model achieved average recall rates of 89.8 % and 75.5 % for pollen classes and Lycopodium , respectively. The classification model correctly categorizes fossil pollen images with >95 % accuracy. We applied the assembled pipeline to Late Glacial pollen samples using class-dependent thresholds to discriminate true pollen from non-pollen objects and compared automated count data for nine pollen types with manual pollen counts. For the selected pollen types, our results demonstrate the feasibility to replicate major fossil pollen changes with automated counts, even when the automated pipeline was applied to pollen samples from a different site than used to train the models. High correlations (r = 0.97) between the first two axes of Principal Component Analyses (PCA) calculated based on automated and manual counts and high correlation (r = 0.93) indicated by a Procrustes rotation analysis of the PCA results demonstrate that the two procedures reconstructed similar pollen patterns. While our automated approach is not yet able to achieve the taxonomic resolution of manual counts by expert analysts and is limited to selected pollen types, it provides a "proof of principle" that automated analyses can be applied to complex fossil pollen samples and to develop downcore stratigraphies. Automated analyses may with time lead to reliable pollen records. For instance, our pipeline can be further improved by adding more pollen classes, increasing the dataset of annotated images of fossil pollen grains, expanding the training data to rare pollen types, refining taxonomic resolution (e.g., separation of Betula nana -type or Pinus -types), and incorporating more challenging pollen types (e.g., Juniperus ), to expand its application beyond reconstructing temporal changes in a few selected pollen types.

Method development and application of object detection and classification to Quaternary fossil pollen sequences

Pollen identification through convolutional neural networks: First application on a full fossil pollen sequence

Classification of bee pollen using visible near infrared spectroscopy

A Computational Approach for Recognition of Electronic Microscope Plant Pollen Images

A user‐friendly method to get automated pollen analysis from environmental samples

A neotropical Miocene pollen database employing image-based search and semantic modeling

Quantitative methods in microscopy to assess pollen viability in different plant taxa

First automatic size measurements for the separation of dwarf birch and tree birch pollen in MIS 6 to MIS 1 records from Northern Germany

How to identify pollen like a palynologist: A prior knowledge-guided deep feature learning for real-world pollen classification

Automated multifocus pollen detection using deep learning

Airborne pollen grain detection from partially labelled data utilising semi-supervised learning

Improving classification of pollen grain images of the POLEN23E dataset through three different applications of deep learning convolutional neural networks

Reconstructing the diversity dynamics of paleo-grasslands using deep learning on superresolution images of fossil Poaceae pollen

AIpollen: An Analytic Website for Pollen Identification Through Convolutional Neural Networks

PollenNet: A novel architecture for high precision pollen grain classification through deep learning and explainable AI

Challenges and opportunities in Quaternary palynology

Pollen recognition through an open-source web-based system: automated particle counting for aerobiological analysis

POLLEN13K: A Large Scale Microscope Pollen Grain Image Dataset

PollenDetect: An Open-Source Pollen Viability Status Recognition System Based on Deep Learning Neural Networks

Comparing whole‐genome shotgun sequencing and DNA metabarcoding approaches for species identification and quantification of pollen species mixtures

Precise automatic classification of 46 different pollen types with convolutional neural networks