Why it matters
Without AI, signals may be misinterpreted, misregistered, or improperly calibrated, risking irreversible material damage, missed underdrawings, and disputed conservation records.
From design to safety
Design
AI-driven imaging proposal design
In the design phase, the system proposes imaging protocols, filter selections, and registration strategies based on material type, expected underdrawing visibility, ethical constraints, and archival data needs, with explicit data schemas and a rationale suitable for accreditation reviews. It specifies file formats (TIFF, RAW, OME-XML), color- and spectral-space conversions, and interoperability with PACR-accredited record systems.
- Infrared camera
- Multispectral camera
- Image registration software
Build
Equipment integration planning
During build, calibration targets and illumination alignment are finalized, hardware-software interfaces established for repeatable data capture, metadata capture, and secure logging, including device IDs, versioning, user permissions, and validation procedures for audit trails. It documents power budgets, rack layouts, shielded cabling, and EMI considerations to support long-term stability.
- Infrared camera
- Illumination system
- Calibration targets
- Infrared filters
Operation
Active imaging and analysis
In operation, AI monitors exposure, noise, alignment, and environmental conditions while producing underdrawing maps, pigment-layer images, and a consolidated report with reference metadata, links to condition records, and treatment proposals, all stored with event-based versioning. Outputs include heatmaps, fluorescence false-color composites, and exportable XML for archival indexing.
- Multispectral camera
- Image registration software
- Calibration targets
Maintenance
Routine calibration and validation
Maintenance includes sensor calibration, filter integrity checks, illumination recalibration, and validation against control samples to sustain data accuracy, with non-conformance logged and addressed by the studio manager, plus quarterly external review of imaging protocols. It specifies calibration interval records, drift thresholds, and replacement part cross-references.
- Infrared camera
- Calibration targets
Safety
Safeguards and documentation
Safety checks verify COSHH-compliant exposure levels, strict no-harm imaging practices, and accumulative logs, with signed treatment-proposal records and controlled access to raw data during analysis, including chain-of-custody for digital outputs. It enforces PPE checks, staff training logs, and ethical review timestamps for every project.
- Illumination system
- Infrared filters
Part by part
- 1
Infrared camera
- Watch for
- Detector drift or thermal noise affecting IR signal quality
- AI action
- AI flags drift, initiates re-calibration and re-acquisition protocol
- 2
Multispectral camera
- Watch for
- Spectral distortion or misalignment between bands
- AI action
- AI detects misalignment, triggers re-registration and band-specific correction
- 3
Infrared filters
- Watch for
- Filter degradation or spectral shift over time
- AI action
- AI alerts to replace or recalibrate filters before data loss
- 4
Illumination system
- Watch for
- Lamp aging or uneven illumination causing hotspots
- AI action
- AI triggers illumination mapping and compensatory exposure adjustments
- 5
Image registration software
- Watch for
- Misregistration across bands or scale drift
- AI action
- AI–driven alignment review and reprocesses datasets
- 6
Calibration targets
- Watch for
- Target deterioration or misalignment during captures
- AI action
- AI schedules retakes with verified targets
Worked example
Detecting hidden underdrawing in a Renaissance panel
Illustrative scenario with sample data.
A private studio analyzes a panel painting with IR and multispectral imaging to reveal underdrawing patterns without removing varnish, supporting a reversible treatment plan and documented provenance.
Signal in
Exposure: IR 850 nm; VIS-IR 700–900 nm; Bands: 3-channel multispectral; Metadata: date, artist, accession, light source, filter IDs, instrument serials, operator, room temperature, humidity, calibration target lot
System out
AI-derived underdrawing map with confidence metrics; multi-layer pigment maps; a treatment report draft with a work-order, material identifiers, and archival metadata references.
Human sign-off: Head of Conservation reviews the underdrawing map, validates imaging protocol, approves treatment proposal, and signs the condition report and archival index entry.
How to roll it out
Phase 1
Weeks 1 to 4
Baseline setup of cameras, filters, software; initial calibration; standard operating procedures established; secure data pathways defined.
Phase 2
Weeks 5 to 12
Pilot imaging on three objects; AI-assisted evaluation of data quality, alignment, and initial underdrawing interpretation; metadata templates refined.
Phase 3
Weeks 13 to 20
Full integration with treatment reports, condition records, and archival metadata; routine use in client projects; formal training completed.
Where people stay in charge
- Decisions on treatment impact, ethics, and candidate materials must be signed by a qualified conservator.
- Imaging protocols and safety parameters must be approved by a conservator before data capture.
- Data handling and privacy must be overseen by the studio's accredited data officer.
Questions owners and crews ask
What is AI used for in infrared reflectography?
AI assists signal interpretation, enhances band alignment, and streamlines reporting; conservators retain final authority for interpretation and treatment decisions.
How is data quality ensured in multispectral analysis?
Calibration targets, exposure controls, and validation against reference samples are used; a conservator signs off on data quality before treatment proposals.
What safety checks accompany intrusive imaging?
COSHH-compliant safety practices are maintained, with protective measures for operators and controlled illumination exposure during data capture.
How are results archived for future access?
All imaging data, metadata, and treatment decisions are stored in a controlled database with versioned reports and access logs.