Designing a camera pilot that produces usable evidence
What should be specified before an indoor farm commissions its first camera-based monitoring pilot?
Scope of this review
A focused desk review of image acquisition documentation, plant phenotyping metadata guidance and an AI evaluation framework. The intended use is a monitored pilot in an existing growing facility. This brief reports no Hamfy field trial or measured model performance.
How we assessed the evidence
We selected primary documentation for three distinct questions: how to acquire interpretable images, how to preserve crop context and how to evaluate a system in its intended use. We translated those principles into a proposed acceptance process. This is a targeted review, not a systematic review or a comparison of camera vendors.
What the evidence supports
Capture is part of the measurement
The PlantCV guidance informed our recommendation to test the imaging arrangement before scaling collection. Our practical extension is to treat a camera move or lighting change as a recorded change in the pilot, with a fresh image-quality check.
Context makes the image interpretable
MIAPPE supports describing the experiment around an observation. We propose a deliberately small operating record: batch, stage, zone, time and interventions. This is an adapted pilot record, not a claim of formal MIAPPE compliance.
Acceptance needs an operating context
The NIST framework informs the requirement for a defined use and ongoing evaluation. Our suggested event grouping, workload measures and independent crop-cycle test are project-design choices to be agreed with the grower.
What this review does not establish
- These sources do not determine camera count, minimum image volume or an acceptable false-alert rate for a particular crop.
- The proposed workflow has not been validated here with images or outcomes from a Hamfy partner site.
- An observation label and a confirmed biological cause require different evidence; the pilot must keep them distinct.
Hamfy’s proposed next step
Before purchasing a full installation, produce one signed-off pilot sheet containing the observable task, capture arrangement, review owner, independent test boundary and stop conditions. Use it to decide whether the proposed data collection can answer the operational question. Change the scope explicitly if the task changes.
Evidence to collect for a real project
- A sample of images across the intended crop stages and lighting states.
- An agreed label definition with examples of ambiguous cases.
- A baseline record of current scouting time and confirmed events.
- An independent test cycle with an auditable review log.
Source notes
- Technical documentation · v4.4Analysis approachesPlantCV documentation
Supports decisions about capture conditions and image analysis. It does not establish the performance of a Hamfy model.
- Research data standard · Accessed 2026Minimum Information About Plant Phenotyping ExperimentsMIAPPE community
Provides a reference for documenting experimental context and making observations interpretable.
- Public framework · 2023AI Risk Management Framework: CoreNIST
Supports context-specific evaluation, responsibility and monitoring. It is not a crop model certification.