A vertically integrated pipeline — drone capture, photogrammetry, cross-modal alignment, deep learning, and geospatial reasoning — delivering decision-ready outputs at site, portfolio, and contractor scale.
Pipeline
01
Acquire
Multi-modal site data capture across asset, portfolio, and contractor scale.
02
Construct
Centimeter-accurate 3D reconstruction with cross-modal co-registration.
03
Recognize
Feature- and defect-level recognition at every cell, every flight.
04
Reason
Domain-aware fault analysis and severity prioritization.
05
Deliver
Geo-located, dispatch-ready work orders and reports.
Differentiated capability
Solar inspections involve two flights — one with an RGB camera, one with a thermal camera. Most platforms treat these as independent datasets, leaving anomalies spatially mismatched between the maps.
Aerometrix solves the alignment at the lowest level. Purpose-built cross-modal feature extractors match individual points between RGB and thermal imagery, then thread those matches into a unified geometric model via a custom bridge-based track architecture that preserves geometric integrity.
The result: every thermal anomaly is geometrically locked to the exact RGB pixel. Repairs go to the right module. Year-on-year degradation trends are comparable. The data compounds.
How it works
Modal-specific feature extraction
Purpose-built feature extractors tuned to the distinct signal properties of optical and thermal imagery.
Bridge-based track construction
Co-temporal RGB ↔ IR observations bridge the modalities; no transitive cross-modal chaining preserves geometric soundness.
Within-modal track extension
Bundle-adjusted tracks extend through descriptor matching and are verified by pose-based triangulation.
Unified orthomosaic output
IR and RGB orthomosaics are co-registered to sub-pixel precision and exported in industry-standard cloud-optimized formats.
Capabilities
Aerometrix unifies photogrammetry, deep learning, and geospatial reasoning in a single production system — from raw flight imagery to ranked, geo-located work orders.
Standard inspections treat RGB and thermal flights as two independent passes. We align them at the tie-point level using purpose-built cross-modal feature extractors, so every anomaly is geometrically locked to the exact pixel across both modalities.
Every module gets a high-dimensional embedding from foundation-model backbones. This powers similarity search across portfolios, unsupervised failure-mode clustering, and few-shot detection of rare defects.
A custom transformer classifier processes every cell with full-panel context via self-attention — not as an isolated crop — capturing fault interactions across the module.
Oriented bounding-box detection of cells, junction boxes, and connectors at sub-cell precision. Resolution-aware tiled inference handles gigapixel orthomosaics with cross-tile non-maximum suppression.
Industrial structure-from-motion produces sub-pixel-accurate orthomosaics, digital elevation models, and dense point clouds. Cross-modal alignment is layered on top for unified IR ↔ RGB co-registration.
A purpose-built thermal converter handles raw radiometric capture with calibrated temperature output. ΔT scoring drives IEC 62446-3 severity classification.
A deterministic rule engine for known fault patterns (substring activation, vegetation shading, burn-mark grouping) is augmented by a reasoning layer for ambiguous cases.
A multi-tenant digital-twin asset model with versioned module replacements and a closed-loop work-order system. Spatial intelligence handles hundreds of thousands of modules per site at sub-second query latency.
Relationship-based access control isolates assets across organizations and contractor relationships. No data leakage despite shared infrastructure.
