Why it matters
Without AI, misread labels, incomplete metadata, and mismatched geolocations can disrupt ex situ conservation, hinder provenance verification, and complicate Index Seminum and accession workflows.
From design to safety
Design
AI-enabled design for digitisation workflow
Define data schema, imaging standards, and metadata fields; map on-sheet elements to database records; specify OCR and NLP models for label types and handwriting; plan georeferencing templates and workflow handoffs to the Herbarium DBMS.
- High-resolution scanner
- Label transcription workstation
- Herbarium database management system
Build
Collect and connect metadata streams
Establish imaging rig setups, calibrate color profiles, implement georeferencing templates, and link to accession records and specimen cabinets; draft data quality checks and versioning rules for records.
- High-resolution scanner
- DSLR imaging rig
- Georeferencing software module
- Herbarium database management system
Operation
Ongoing capture and verification
Capture sheet images, transcribe labels, verify provenance, attach geospatial coordinates, and update living-collections and seed-conservation records; perform routine reconciliation with Index Seminum and physical cabinets.
- DSLR imaging rig
- Label transcription workstation
- Georeferencing software module
- Herbarium database management system
Maintenance
Quality control and data integrity
Run regular OCR accuracy checks, audit metadata fields, back up records, recalibrate imaging devices, and perform preventive maintenance on scanners to maintain consistent data quality over time.
- Label transcription workstation
- High-resolution scanner
Safety
biosecurity and data governance
Protect sensitive provenance data, enforce tiered access controls, and document handling policies for fragile sheets; ensure legal provenance is preserved during digitisation workflows.
- Herbarium database management system
- Georeferencing software module
Part by part
- 1
High-resolution scanner
- Watch for
- Scan quality degradation or bed misalignment causing partial captures over multiple sheets.
- AI action
- Detect cut-off edges and trigger re-scan order with alignment check.
- 2
DSLR imaging rig
- Watch for
- Inconsistent lighting or color drift across sessions affecting image color fidelity.
- AI action
- Adjust white balance and lighting presets; flag exposures needing retake.
- 3
Label transcription workstation
- Watch for
- Handwritten fonts and faded ink produce transcription errors.
- AI action
- Run OCR/NLP models, surface uncertain fields for manual review.
- 4
Georeferencing software module
- Watch for
- Mismatched coordinates with established collection site data.
- AI action
- Compute centroid coordinates, flag anomalies for curator check.
- 5
Herbarium database management system
- Watch for
- Inconsistent accession links or missing provenance data.
- AI action
- Resolve by cross-referencing with Index Seminum and accession records.
- 6
Specimen cabinet and associated metadata rails
- Watch for
- Physical misplacement or mislabeling of sheets during handling.
- AI action
- Log movement; verify sheet-id against digital record.
Worked example
Digitising a mid-sized garden’s herbarium sheets to align 12,000 records
Illustrative scenario with sample data.
A mid-sized garden with 12,000 herbarium sheets seeks comprehensive digitisation, improved metadata, and linked provenance to living collections, seed bank, and conservation science workflows.
Signal in
sheet_id: H-2024-0012, label_text: “Nicotiana tabacum, exs. 1890, provenance: Peru”; date_imaged: 2025-08-12; georef_source: old map; image_resolution: 600 dpi; OCR_confidence: mixed; collector: “A. Rivera”; habitat: “savanna edge”
System out
record_link: accession_id A-12345; metadata: {species: Nicotiana tabacum, origin: Peru, collection_year: 1890, verified: true, collector: A. Rivera, habitat: savanna edge}; georeference: lat 12.3456, lon -78.9876; transcription_review: required; provenance_note: “Index Seminum 1920, updated 2024”
Human sign-off: Herbarium curator reviews uncertain labels, approves final linkage to accession records, and signs off on provenance notes in the Index Seminum record.
How to roll it out
Phase 1
Phase 1
Baseline digitisation of 5,000 sheets, establish imaging and transcription pipelines, validate initial metadata mappings, and define data governance rules for the herbarium DBMS.
Phase 2
Phase 2
Scale to 12,000 sheets, implement georeferencing checks, integrate with living-collections records, and harmonise with seed-bank accession schemas.
Phase 3
Phase 3
Full operational digitisation with ongoing quality control, archival backups, and periodic audits of provenance integrity and data provenance lineage.
Where people stay in charge
- Decisions on data governance, access controls, and provenance verification must be signed off by the Herbarium Director or Plant Records Manager.
- Georeferencing and image resolution standards must be approved by the Plant Records Manager and Head of Conservation Science.
- All handling policies for fragile sheets and legal provenance must be reviewed and signed by the Conservation Officer before digitisation begins.
Questions owners and crews ask
What are the core data elements AI must capture from herbarium sheets?
The AI must extract species name, collection date, collector, locality, habitat notes, accession provenance, and any conservation status notes, aligning each with the database schema and existing records.
How does AI handle handwriting and vintage labels?
AI uses OCR and NLP models trained on historical handwriting, flags uncertain fields for human review, and records confidence metrics alongside transcribed text to preserve traceability.
How is georeferencing integrated with living collections data?
Georeferenced sheet coordinates are stored in the herbarium DBMS and cross-checked against seed bank and field collection locations to ensure provenance consistency across records.
What safeguards protect fragile sheets during digitisation?
Handling guidelines are followed, imaging requires trained staff, and high-resolution scans are produced with noninvasive methods and cabinet-level containment to minimize disturbance.