TDWG2026
Home / Talks / SAL C
talk · Tuesday 22 September · 14:45–⁠15:00 · SAL C

X-raying wild plant seeds: preparing images for future AI-driven global biodiversity research

Robert Turner · Digitizing Collections & Legacy Data

Recording time 6:03:39–6:17:15Open on Vimeo ↗

Summaries and key moments are machine-produced. Check the recording for the speaker’s exact words.

The short versionAdding scale and linking legacy seed X-rays to collection records turns an inaccessible archive into AI-ready research data.

Overview

What this was about

Robert Turner explained how Kew is migrating 21 years of Millennium Seed Bank seed X-ray images (used for non-destructive seed quality testing: healthy, empty or infested seeds) from a network drive into its digital asset management system and linking them to collections data. As part of Kew's 2021-2026 digitisation programme and new Integrated Collections Management System (Earthscape, occurrence-centred with links to WCVP/POWO and Index Fungorum), Python tools add a resolution-based grid and scale bar to each image, validated against physical seed measurements. Images go into the Digifolia DAMS, are exposed via S3/CloudFront URLs linked in Earthscape, and will reach users through the Kew data portal and the future Seed Pod portal, subject to checking material and data agreements.

Why it matters. Seed X-rays could support AI research on identification, viability and damage across the world's largest ex situ seed collection.

Key ideas

In the room

  • X-raying is non-destructive, rapid and suitable for small seeds, unlike destructive cut tests.
  • The Millennium Seed Bank holds about 105,000 seed collections of 40,000 species from 191 countries.
  • 21 years of X-ray images were stored on a network drive inaccessible to most staff and external scientists.
  • Potential AI uses include taxonomic identification, morphology, germination/viability and infestation/damage detection.
  • Python tools read DPI from image metadata or DICOM files to compute grid spacing and add scale labels.
  • Earthscape separates collection objects from occurrences so herbarium, seed and living collections from one event share an occurrence.
  • Data sharing is constrained by material agreements with countries whose alignment with data agreements must be checked.
Jump in

Notable moments

In their words

Transcript

Automatically generated captions can contain mistakes, especially in names and technical terms. Times are relative to the room recording.

Read the transcript ↓
Loading transcript…
Keep exploring

Related talks.

See the map ↗
talk · 24 Sept · 16:38–⁠16:50 · FORUM

Digitising Plant and Fungal Collections at Royal Botanic Gardens, Kew: Applying International Standards to Support Data Sharing and Geospatial Analysis

WGSRPD enables cross-collection geospatial analysis at Kew, but historical, heterogeneous data leave half of records without fine-level assignment.

Udayangani Liu
talk · 22 Sept · 15:00–⁠15:15 · SAL C

Next-generation digitization: integrating spectral reflectance into the online mobilization of herbaria

Spectral reflectance can be captured at scale during herbarium digitisation and supports accurate species identification from leaves alone.

Matthew Austin
talk · 22 Sept · 10:05–⁠10:20 · SAL C

Understanding geological specimen use for effective digitisation strategies

Evidence from inquiries, loans and publications shows geological specimens have diverse, sometimes unexpected uses, which should shape which data are digitised first.

Rachel Walcott, Adam Mansur
Programme abstract

X-raying wild plant seeds: Preparing images for future AI-driven global biodiversity research

Robert Turner

X-ray imaging is an effective non-destructive method for assessing seed quality in seed banks, enabling the identification of full, empty or damaged seeds when estimating the proportion of potentially viable seeds within a sample. This method is routinely used at the Millennium Seed Bank, RBG, Kew, as an alternative to destructive cut-testing for evaluating seed lot quality. Over the past 25 years, more than 100,000 seed x-ray images have been captured, providing a valuable resource for documenting seed traits in wild plants (e.g., shape, size, length, width, and internal features). Here, we describe the processing of these images to generate a corresponding image with a structured grid overlay and standardised measuring scale, to support future AI-driven research to measure features. The image-processing script is Python-based and uses Pillow for standard image manipulation and pydicom for parsing medical-grade X-ray metadata. Script converts digital pixel data into a calibrated physical reference through a three-step mathematical transformation: (1) metadata extraction identifies image resolution from file headers by reading dots per Inch (DPI); for standard JPEGs, this is obtained via Pillow, while for DICOM files, resolution is derived from pixel spacing or equivalent tags using pydicom; (2) pixel-to-metric conversion computes pixels per centimetre using standard conversions, enabling grid intervals according to image resolution; and (3) grid placement generates an evenly spaced coordinated sequence across image dimensions. Scale accuracy is validated against physical seed length measurements from a representative sample. This workflow standardises legacy and newly acquired x-ray images, enabling downstream computational and AI-driven global biodiversity research.

Submitted to the conference programme before the talk, so the talk itself may differ.