In palynology, the visual classification of pollen grains from different species is a hard task which is usually tackled by human operators using microscopes. Many industries, including medical and pharmaceutical, rely on the accuracy of this manual classification process, which is reported to be around 67%. In this paper, we propose a new method to automatically classify pollen grains using deep learning techniques that improve the correct classification rates in images not previously seen by the models. Our proposal manages to properly classify up to 98% of the examples from a dataset with 46 different classes of pollen grains, produced by the Classifynder classification system. This is an unprecedented result which surpasses all previous attempts both in accuracy and number and difficulty of taxa under consideration, which include types previously considered as indistinguishable.

In palynology, the visual classification of pollen grains from different species is a hard task which is usually tackled by human operators using microscopes. Its complete automatization would save a high quantity of resources and provide valuable improvements especially for allergy-related information systems, but also for other application fields as paleoclimate reconstruction, quality control of honey based products, collection of evidences in criminal investigations or fabric dating and tracking. This paper presents three state-of-the-art deep learning classification methods applied to the recently published POLEN23E image dataset. The three methods make use of convolutional neural networks: the first one is strictly based on the idea of transfer learning, the second one is based on feature extraction and the third one represents a hybrid approach, combining transfer learning and feature extraction. The results from the three methods are indeed very good, reaching over 97% correct classification rates in images not previously seen by the models, where other authors reported around 70.

In palynology, the visual classification of pollen grains from different species is a hard task which is usually tackled by human operators using microscopes. Many industries, including medical and farmaceutical, rely on the accuracy of this manual classification process, which is reported to be around 67%. In this paper, we propose a new method to automatically classify pollen grains using deep learning techniques that improve the correct classification rates in images not previously seen by the models. Our proposal manages to properly classify up to 98% of the examples from a dataset with 46 different classes of pollen grains, produced by the Classifynder classification system. This is an unprecedented result which surpasses all previous attempts both in accuracy and number and difficulty of taxa under consideration, which include types previously considered as indistinguishable. 2 scientific fields. An estimated 40% of the world's population experience seasonal allergic 3 rhinitis (SAR) driven by exposure to pollen [1]. Pollen forecasting, informed by 4 examination of airborne pollen has become a key tool for management of SAR [2]. 5 Pollen is also very important for quality verification of honey [3], reconstructing past 6 vegetation to understand past changes in climate change [4], biodiversity [5], and human 7 impacts [6] and as a forensic tool [7]. Common to all these areas is the need for 8 experienced analysts to spend considerable amounts of time identifying and counting 9 pollen on slides. While other branches of science have been transformed by the 10 technological advances of recent decades, palynology is languishing, with the practical 11 methodology of pollen counting having hardly advanced much beyond that of the 1950s. 12 But this is not for want of trying. Flenley [8] was the first to call attention to the 13 need and potential of automation of pollen counting. A handful of early attempts were 14 published in the later decades of the 20th century, but the rapid increase in capability 15 in computational intelligence over the early part of the 21st century resulted in 16 considerable acceleration in the field during this time, with numerous attempts at 17 partial or complete automation of palynology appearing in the literature, summarised 18 in [9] and [10]. 19While the results of the existing studies can be regarded as promising, they are 20 rather limited in that they typically deal with a relatively small number of taxa (max 21 February 5, 2020 1/14 30, mean 8), and success/accuracy rates vary. While some palynological applications 22 may require a lower level of taxonomic diversity than others, it is arguable that many 23 'real world' pollen problems will require higher diversity than that of most of the 24 existing studies. For example, Stillman and Flenley, (1996) suggested that the minimum 25 number of taxa for paleoecological applications would be around 40 types. 26Recently, Sevillano and Aznarte ( [10]) presented an example of pollen classification 27 which applied deep learning convolu...