MODEL-FIRST CV
RGB image→Pretrained model→Inference
Effective for standard visual tasks when the imagery already contains a strong, usable signal.
Measurement-first vision engineering
Intelligence begins before the first pixel.
I engineer the full path from physical signal to operational decision, spanning illumination, spectrum, optics, sensors, radiometry, spatial context, models, and deployed inference.
01 / THE DIFFERENCE
A better model cannot recover information the sensor never captured.
Most computer-vision projects begin with an available RGB image and a pretrained model. I work farther upstream, determining what must be observed, how it should be measured, and whether the resulting signal can support a reliable operational decision.
02 / MEASUREMENT BEFORE MODEL
Computer vision does not begin with the model.
It begins with the physical phenomenon and the information the sensing system can preserve.
MEASUREMENT-FIRST VISION
Phenomenon→Sensor system→Information→Inference
Designed for problems where wavelength, resolution, atmosphere, geometry, calibration, or deployment conditions determine whether the answer is observable at all.
Before choosing a model, I ask:
- Is the target physically observable?
- At which wavelengths?
- At what spatial and temporal resolution?
- Under what illumination conditions?
- During what atmospheric conditions?
- Can the sensing system preserve the required signal?
03 / PHYSICS TO DECISION
The complete sensing and inference chain.
Every stage changes what the system can know. Treating the chain as one engineering problem prevents a polished model from being built on weak or misleading measurements.
- 01PhysicsMaterial · Temperature · Motion · Biological state
- 02Illumination & SpectrumReflection · Emission · Absorption · Atmosphere
- 03Sensor & OpticsModality · Bands · Resolution · Viewing geometry
- 04Radiometry & PixelsCalibration · Sampling · Noise · Dynamic range
- 05Spatial & Spectral InformationFeatures · Registration · Context · Geolocation
- 06Model & InferenceDetect · Segment · Classify · Track · Estimate
- 07Operational DecisionAlert · Prioritize · Automate · Act
04 / BEYOND RGB
The world contains more information than human vision can see.
RGB records three broad channels designed around human perception. Many valuable physical phenomena reveal themselves through spectral, thermal, temporal, geometric, or geospatial information that ordinary imagery does not preserve.
- VegetationSpectral response and temporal change
- Fire & SmokeThermal and multispectral evidence
- Material IDAbsorption and reflectance characteristics
- MaritimeSpatial context, atmosphere, and sensor fusion
05 / THE METHOD
The same path, every project.
Each system starts with what can be observed, then follows the signal through sensing and interpretation to an operational decision. Pick an example to see how each step was decided.
06 / SELECTED SYSTEMS
The approach, demonstrated.
A selection of systems across maritime, wildfire, agriculture, and defense. Open any project for the problem, sensing approach, processing steps, and results.
07 / WHAT I BUILD
Observe. Measure. Interpret. Deploy.
Observe
Determine how the physical phenomenon becomes visible and select the appropriate sensing modality, wavelengths, optics, resolution, and acquisition conditions.
Measure
Preserve physical meaning through radiometric and geometric calibration, registration, signal-quality analysis, and spatial context.
Interpret
Apply spectral and spatial features, classical vision, foundation models, detection, segmentation, classification, tracking, and sensor fusion where each fits.
Deploy
Connect inference to cameras, embedded hardware, geospatial pipelines, edge-to-cloud architecture, APIs, and operational workflows.
08 / WHERE IT WORKS
Built around the operating problem.
StartupsDefense & GovernmentMaritimeAgriculture & EnvironmentSmart Cities & Infrastructure
09 / GLOBAL REACH
Field-tested across continents.
Sensing problems change with latitude, climate, and terrain. Projects span coastal, agricultural, urban, and remote environments.
Organizations I’ve worked with
Logos are trademarks of their respective owners. Their display reflects past or current project work and does not imply endorsement.
10 / FOUNDER
Built by an engineer, not an agency.
Spectral Mapping is the independent consultancy of Bernard Lawes, a vision and geospatial AI engineer working across the complete chain from physical signal to deployed inference.
Clients work directly with the person architecting and prototyping the solution, from early technical discovery through proof of concept and deployment planning.
Contact Bernard directly11 / START A CONVERSATION
Is the problem observable, measurable, and solvable?
Let’s evaluate the physical signal, sensing constraints, available data, and smallest credible prototype before committing to a model.
Discuss a prototype hello@spectralmap.ioor send a message here