Leveraging Phytolith Research using Artificial Intelligence
Authors: Andrés G. Mejía Ramón, Kate Dudgeon, Nina Witteveen, Dolores Piperno, Michael Kloster, Luigi Palopoli, Mónica Moraes R., José M. Capriles, Umberto Lombardo
Organizations: Institut de Ciència i Tecnologia Ambientals, Universitat Autònoma de Barcelona Carrer de les Columnes s/n, Cerdanyola del Vallès, Barcelona, 08193, Espanya · Department of Anthropology, Smithsonian National Museum of Natural History, 1000 Madison Drive NW, Washington, D.C., 20560, United States · Smithsonian Tropical Research Institute, Luis Clement Avenue, Bldg. 401 Tupper, Ancón, Ciudad de Panamá, República de Panamá · Phycology Group, Faculty of Biology, University of Duisburg-Essen, Universitätsstr. 2, 45141 Essen, Deutschland · Dipartimento di Ingegneria e Scienza dell’Informazione, Università di Trento Via Sommarive 9, 38123, Trento, Italia · Herbario Nacional de Bolivia, Instituto de Ecología, Universidad Mayor de San Andrés, Calle 27 y Andrés Bello s/n, Cota Cota, La Paz, Bolivia · Department of Anthropology, The Pennsylvania State University, Susan Welch Liberal Arts Building, University Park, Pennsylvania, 16803, United States · Departament de Prehistòria, Universitat Autònoma de Barcelona, Edifici B, Carrer de la Fortuna, Cerdanyola del Vallès, Barcelona, 08193, Espanya
Phytolith analysis is a crucial tool for reconstructing past vegetation and human activities, but traditional methods are severely limited by labour-intensive, time-consuming manual microscopy. To address this bottleneck, we present Sorometry: a comprehensive end-to-end artificial intelligence pipeline for the high-throughput digitisation, inference, and interpretation of phytoliths. Our workflow processes z-stacked optical microscope scans to automatically generate synchronised 2D orthoimages and 3D point clouds of individual microscopic particles. We developed a multimodal fusion model that combines ConvNeXt for 2D image analysis and PointNet++ for 3D point cloud analysis, supported by a graphical user interface for expert annotation and review. Tested on reference collections and archaeological samples from the Bolivian Amazon, our fusion model achieved a global classification accuracy of 77.9% across 24 diagnostic morphotypes and 84.5% for segmentation quality. Crucially, the integration of 3D data proved essential for distinguishing complex morphotypes (such as grass silica short cell phytoliths) whose diagnostic features are often obscured by their orientation in 2D projections. Beyond individual object classification, Sorometry incorporates Bayesian finite mixture modelling to predict overall plant source contributions at the assemblage level, successfully identifying specific plants like maize and palms in complex mixed samples. This integrated platform transforms phytolith research into an "omics"-scale discipline, dramatically expanding analytical capacity, standardising expert judgements, and enabling reproducible, population-level characterisations of archaeological and paleoecological assemblages.
Figures & tables
Figure 1: Sorometry workflow pipeline, with associated data productsin blue.
Figure 2: Sorometry Graphical User Interface.
Figure 3: Data pipeline architecture, with the 2D image stream in green, the 3D pointcloud stream in blue, and fusion modality in purple.
Figure 4: Two representations of the same Cyperaceae Polygonal Achene Body phytolith: (Left) 2-Dimensional RGB Orthoimage crop; (Right) 3-Dimensional Pointcloud
Figure 5: Digital images of micro-fossils included in the study. (a) Acute bulbosus , ACUBUL; (b) Blocky , BLO; (c) Bulliform flabellate , BULFLA; (d) Elongate entire , ELOENT; (e) Elongate sinuate , ELOSIN; (f) Elongate dendritic/dentate, ELODET/DEN; (g) Bilobate , BIL; (h) Cross , CRO; (i) Rondel , RON; (j) Saddle , SAD; (k) Decorated tabular, TABDEC; (l) Polygonal achene body , POLACH; (m) Scalloped sphere, SSP; (n) Trough , TRO; (o) Spheroid echinate , SPHECH; (p) Spheroid symmetrical, SPHSYM; (q) Ellipsoidal echinate , ELLECH; (r) Druse , DRU; (s) Non-phytolith Sponge spicule , SPO; (t) Non-phytolith Silica Microsphere , SilicaBall . Scale bar: 10 µm. The red overlay represents the automatically segmented portion of the phytolith based on the point-cloud extraction. Detailed descriptions of morphological criteria, sub-categories of diverse groups and rationale for inclusion in the study are provided in Supplementary Materials B.
Figure 6: Morphotype classification performance for the Fusion model, showing percent true values classified as a given predicted value.
Figure 7: Morphological classification performance for the Fusion model, showing percent true values classified as a given predicted value.
Figure 8: Morphological classification performance for the Fusion model on well-segmented phytoliths against an archaeological validation slide, showing percent true values classified as a given predicted value.
Figure 9: Composition diagram for volume UABPL-240125d. (Left) Bar graphs displaying the predicted number of objects for each category in each model. (Right) Kernel density estimate of the percent certainty for all objects given their maximum probability class.
Figure 10: ‘Barcode’ diagram, used to visually the composition of a set of slides. Colours correspond to the same key as Figure 9 .
Figure 11: Chi square test of independence for slides by sector. Sectors with significant chi square results have highlighted loadings in green.
Figure 12: Position of each slide in compositional space according to the predicted morphological compositions.
Figure 13: Hierarchical clustering analysis of all slides analysed
Figure 14: Results of the mixture models on a prepared slides of two palm species, Attalea sp. and Euterpe sp., with the model correctly identifying the correct two palms as the most likely contributors
Figure 15: Results of the mixture models on a prepared slides of maize and a bamboo species, with the model correctly identifying maize cob and Chusquea sp. as the two most probable contributors.
Figure 16: Results of the mixture models on a prepared slides of maize, a palm species, a bamboo species, and a plant not in the reference set. Of the six most probable plants identified by the model despite the Strilitzia compound, the first two are maize, three and five are palms, and six is bamboo.
Computer methods that learn patterns from examples and then use those learned patterns to make predictions on new, unseen data. In this study, AI is used to help identify and summarize phytoliths automatically.
Sector
A 2 mm × 2 mm sampled volume of a slide that is scanned and processed as one unit. Multiple sectors can be scanned from a single slide so that the slide can be sampled efficiently without always digitizing the entire slide.
z-stack (focus stack)
A series of images of the same sector taken at many different focal depths. Together, these images record how objects appear as the microscope focus moves through the settled material on the slide.
z-slice
One individual image plane within a z-stack, corresponding to a single focal depth.
Focus-stacked orthoimage (orthomosaic)
A single two-dimensional image made by combining the sharpest pixels from all z-slices, so that as much of the sector as possible appears in focus at once.
Point cloud
A three-dimensional representation of an object or surface made up of many points with x, y, and z coordinates, often also carrying RGB color values. In Sorometry, point clouds approximate the 3D form of segmented microscopic objects.
Table S.A1: Plain-language definitions of key technical terms used in the Sorometry workflow.
Object Type
Definition
Satisfactory phytolith
Single, well-segmented phytoliths or other objects where all or most key morphological features are captured in the point cloud and segmentation. Some low-level additional detritus or noise is acceptable.
Fragmentary phytolith
Only part of the phytolith is captured in the point cloud and segmentation, or part of the phytolith falls out of the field of view. Some diagnostic features are still captured.
Poorly-segmented phytoliths
More than one discrete object or phytolith is captured in the segmentation.
Multicell
Two or more phytoliths joined in anatomical connection, or phytoliths with other parts of the silica structure attached.
Trash
Segmentation detritus; objects capturing only undiagnostic elements of a whole phytolith (e.g. only one edge); small non-phytolith objects ( <5μ m); other non-phytolith material (e.g. quartz, clay aggregates); silica fragments; and segmented objects with such low image and point-cloud resolution that they cannot be securely identified as phytoliths.
Sponge
Sponge spicule .
Table S.B1: Definitions of object types used in classification.
Group
Total Count
Multicell
281
PoorlySegmented
3060
Singlet
8859
SpongeSpicule
68
Trash
771
Table S.B2: Segmentation Quality and Object Type Count (Train + Test)
Group
Total Count
Multicell
8
PoorlySegmented
260
Singlet
1402
SpongeSpicule
9
Trash
521
Table S.B3: Segmentation Quality and Object Type UABPL-240151d (Validation Set)
Group
Total Count
AcheneBody
17
AcuBul1
174
AcuBul4
116
AcuBulOther
183
Bilobate
344
Blocky
301
Table S.B4: Morphotype Count (Train + Test)
Group
Total Count
AcuBul4
10
AcuBulOther
28
Bilobate
15
Blocky
51
BulFlab
12
Cross
3
Table S.B5: Morphotype Validation Set from UABPL-240151d
Code
Name
Name in confusion matrix
Key morphological attributes
ACUBUL
Acute bulbosus
AcuBul1 ; AcuBul4 ; AcuBulOther
Narrow body with acute apex and wider antapex. Size between ∼ 25–100 μ m
BLO
Blocky
Blocky
Solid 3D bodies with length/width <2
BULFLA
Bulliformflabellate
BulFlab
Well silicified “fan shape”, lower part narrower than upper part
ELOENT
Elongate entire
EloEnt1 ; EloEnt2 ; EloEnt3
Rectilinear 2D outline with predominately entire (smooth) margins. L:W >2
ELOSIN
Elongate sinuate
EloSin
Rectilinear 2D outline with sinuate margins (alternating concavities and convexities). L:W >2
ELODET/DEN
Elongate dentate/dendritic
EloDetDen
Rectilinear 2D outline with dentate margins (acute projections), and occasionally branched processes. L:W >2
Table S.B6: Key phytolith morphotype groups included in the training set, with corresponding names used in the confusion matrix. Descriptions are based on ICPN 2.0 unless otherwise stated.
Slide &
Count
Context
Sector ID
Type
(Morphotype/Quality)
FI 757 [10] L1089 S399
UABPL-240157a
ARCH
232 / 1405
UABPL-240157b
ARCH
208 / 1041
UABPL-240157c
ARCH
311 / 869
FI 757 [27] L1094 S448
UABPL-240161a
ARCH
1 / 35
FI 757 [39] L1088 S419
UABPL-240151b
ARCH
3 / 4
Table S.B7: Provenance of scanned microscopic objects used in this study.
Figure S.D1: Segmentation quality and object type classification performance for the ConvNeXt 2D image model, showing percent true values classified as a given predicted value.
Figure S.D2: Segmentation quality and object type classification performance for the PointNet++ 3D pointcloud model with late fusion scalar, showing percent true values classified as a given predicted value.
Figure S.D3: Morphotype classification performance for the ConvNeXt 2D image model, showing percent true values classified as a given predicted value.
Figure S.D4: Morphotype classification performance for the Pointnet++ 3D pointcloud model with late fusion scalar, showing percent true values classified as a given predicted value.
Figure S.E1: Learning Curve for the ConvNeXt 2D image morphological model (Classified) and segmentation quality/object type model (AllClasses).
Figure S.E2: Learning Curve for the Pointnet++ 3D pointcloud morphological model (Classified) and segmentation quality/object type model (AllClasses).
Figure S.E3: Learning Curve for the 2D-3D Fusion morphological model (Classified) and segmentation quality/object type model (AllClasses).
Palynomorphs (microscopic, organic-walled fossils such as pollen, spores, and dinoflagellates) are important high-resolution records of past climates and are critical to the study of ancient ecosystems. Existing methods rely on manual analysis of high-resolution, multifocal digital microscopy images, which is slow and time-consuming and requires researchers to compromise on the scale of their investigations. To the best of our knowledge, our work proposes the first ever scalable end-to-end pipeline for automated palynomorph detection in whole slide images that addresses this bottleneck through: (1) efficient methods for decomposing and compressing digitized multifocal microscope slide images into tractable 2-dimensional tiles for analysis; (2) benchmarking modern object detection models, including RF-DETR, for the detection of palynomorphs, achieving an AP@50 of 0.879; (3) an efficient algorithm for the synthesis of detection outputs across large-scale, high-resolution images; and (4) an I/O optimization resulting in faster inference time. Our methods drastically reduce the time required for palynomorph detection in a single slide from often days of manual inspection to under one hour of automated analysis, enabling palynological research at a substantially greater scale.
Abbas Shaikh, Praise Mayor, Patrick Ainlay-Vazquez +7
Rice University Houston TX USA · Smithsonian National Museum of Natural History Washington DC USA · Smithsonian Office of Digital and Innovation Washington DC USA
The study of ceramic materials constitutes a cornerstone of archaeological research, yet the post-production workflow for pottery documentation remains labor-intensive and creates significant publication bottlenecks. This paper presents PyPottery, an open-source, AI-powered suite designed to semi-automate the complete ceramic documentation pipeline. The suite comprises four integrated modules: PyPotteryScan for automated image extraction and handwriting recognition; PyPotteryInk for automatic inking of pencil drawings; PyPotteryTrace for semantically-aware vectorization; and PyPotteryLayout for automated layout generation. Evaluated on 50 hand-drawn sheets containing 240 pottery drawings from the Terramara di Montale (Italy), the framework achieved substantial time savings confirmed by usability study participants, who reported a median perceived speedup of 40× over traditional workflows (range: 17.5×--120×). These results highlight the potential of AI-assisted tools in archaeological documentation, while the paper addresses the strategic redistribution of cognitive labor toward augmentation rather than automation.
Lorenzo Cardarelli
Seminars für Ur- und Frühgeschichte Georg-August-Universität Göttingen Göttingen, 37073 · ISMed-CNR – Istituto di Studi sul Mediterraneo Naples, 80134 · Sapienza Università di Roma, Department of Antiquities, and Parco Archeologico del Colosseo, Via in Miranda 5, 00186 Rome, Italy
Traditional melissopalynology is a time-consuming and subjective process, often taking 4-6 hours per sample. We present an automated, high-throughput microscopy system that integrates H_\infty robust mechanical control with advanced deep learning pipelines for the precise counting, classification, and morphological analysis of pollen grains from Bio Bio region in south central territory in Chile. Our system employs U^2-Net for salient object detection and a DINOv2 Vision Transformer backbone trained via Deep Metric Learning for classification. By integrating Gradient-Weighted Attention, the model provides human-interpretable texture and diagnostic feature annotations. The system achieves a 95.8% classification recall and at least 6x processing speedup compared to manual expert analysis.
J. Staforelli-Vivanco, R. Jofré, P. Coelho +12
Facultad de Ingeniería, Universidad Andres Bello, Talcahuano, CCP-THNO 7100 Chile · Universidad Arturo Prat, Sede Victoria, Chile · Laboratory of Palynology and Plant Ecology, School of Sciences and Technologies, University of Concepción, Los Angeles Campus