Bands and compositions
Each band records a different spectral response from the surface.
B / G / R / RE / NIR / SWIR— Nexa Vision
Nexa Vision: computer vision applied to AI.
Nexa Vision gives Convex's robots eyes. Perception, scene understanding, diagnosis and reporting — from sensor to management dashboard, in a single pipeline.
The robot's vision layer, between the pixel and the decision.
Autonomous robots are only useful when they see the world with the same care as an experienced operator. Nexa Vision combines detection, segmentation, classification and multimodal models to turn raw imagery into structured scene understanding.
This layer feeds both the navigation and actuation of Convex robots — like Axis and Synapse Control-based drones — and the diagnosis and recommendation reports delivered to the grower.
Analytical flow
From image to understanding.
Nexa Vision turns visual and spectral signals into operational interpretation.
- 01
Capture
Orbital and aerial records of the same area, across multiple bands.
Satellite / drone / optical sensor - 02
Processing
Images are aligned, corrected and clipped to the field boundary.
Bands / composition / normalization / clipping - 03
Index extraction
Band combinations become indicators that stay comparable over time.
NDVI / NDMI / GNDVI / SAVI / NBR - 04
Visual reading
Within-field contrast reveals patterns, failures and anomalies.
Patterns / anomalies / priority zones - 05
Interpretation
Each reading gains agronomic context and operational history.
Time series / passes / comparison - 06
Report
Findings organized into a technical summary and recommendation.
Summary / recommendations / communication
CaptureNexa VisionReport
Nexa Vision
A general-purpose computer vision model that interprets different types of imagery to support decisions in agriculture, industry, mining, logistics, security, livestock farming, oil and gas, and environmental management.
Nexa Vision's pillars
Perception
Detection, segmentation and classification of objects, surfaces and patterns in images captured by drones, mobile robots and fixed field cameras.
Scene understanding
Multimodal models understand context: field, crop, growth stage, terrain type, infrastructure condition and the presence of animals or people.
Robotic decision
Structured outputs feed the navigation, actuation and task-prioritization logic of Convex robots and third-party systems.
Diagnosis and report
Every mission generates a visual report with detections, indicators and recommendations, ready for the technical team and the manager.
How the Nexa Vision pipeline works, from capture to action
Capture
Images from drones, Axis robots, fixed cameras or the field team's smartphone.
Pre-processing
Geometric correction, mosaicking, georeferencing and sync with robot telemetry.
Inference
Computer vision and multimodal models run in the cloud or at the edge, close to the robot.
Fusion
Detections are cross-referenced with maps, history and agronomic or operational rules.
Action
The result becomes a command for the robot, an alert for the operator, or a report for the manager.
Observed signals
What the system observes.
Four complementary readings of the same area.
Spectral indices
Band combinations quantify vigor, moisture and degradation.
NDVI / NDMI / GNDVI / NBRSpatial variation
Within-field contrast separates zones that behave differently.
Zones / edge / failure / patchTemporal evolution
Repeated readings across passes show trend and persistence.
T1 → T5 / series / trend— Report cycle
From satellite to report, one cycle per valid observation.
Every Nexa Vision report comes from a new valid Sentinel-2 optical observation. With no usable scene, no report is simulated: the cycle waits for the next eligible pass.
- 01Sentinel-2
- 02new pass
- 03quality control
- 04valid observation
- 05processing
- 06indices
- 07historical comparison
- 08Nexa M1
- 09PDF
- 10notification
Report frequency
- —Weekly report
- —Every 10 days
- —Every 15 days
- —Every 30 days
Weekly report
Standard frequency of one report every 7 days. The image used depends on the availability of a new valid optical observation.
When no usable scene is available
No suitable new optical observation was available for this cycle. Monitoring will continue automatically.
Nexa M1 included
AI interpretation included with each completed analysis, subject to a fair-use policy: each completed analysis includes an allowance of up to 1,000 Nexa M1 output tokens for interpreting that analysis.
Report language: Brazilian accounts receive the PDF in Portuguese by default and international accounts in English; a manual language preference can override that.
Analysis architecture
Layers that organize the visual reading.
Each layer prepares, qualifies and communicates the output of the previous one.
01Optical / multispectral / orbital / aerial
Visual input
Satellite, drone and optical sensors record the same area at different resolutions and bands.
02Alignment / normalization / indices
Spectral processing
Bands are aligned, normalized and combined into indices that answer specific questions.
03Segmentation / patterns / context
Visual and contextual interpretation
Computer vision models read spatial variation and anomalies with agronomic and historical context.
04Maps / summary / recommendation
Technical and operational communication
Findings are structured into maps, summaries and recommendations the team can act on.
Conceptual scheme · schematic reading, not to scale
Multiple lenses to read vegetation and terrain.
Nexa Vision processes multispectral images and generates indices that quantify vigor, moisture, chlorophyll, surface water and degraded areas. Each index answers a specific operational question and can be combined for more robust diagnoses.
NDVI
Vegetation vigor and density. The basis for zoning and gap detection.
Ranges
Bare soil / gapLow vigorModerate vigorGood vigorExcellent vigor
NDRE
Red-edge response, sensitive to chlorophyll and nitrogen at advanced growth stages.
Ranges
Very lowLowModerateHighVery high
GNDVI
Vigor computed on the green band, useful under denser canopies.
Ranges
Very lowLowModerateHighVery high
SAVI
Soil-adjusted vigor, highlighting areas where exposed soil skews the reading.
Ranges
Soil dominantLow coverModerate coverHigh coverVery high cover
EVI
Vigor corrected for atmosphere and soil, with better response in high biomass.
Ranges
Very lowLowModerateHighVery high
NDMI
Vegetation moisture. Indicates possible canopy water stress.
Ranges
Water deficitLow moistureAdequate moistureHigh moisture
NDWI
Surface water and waterlogging within the area boundaries.
Ranges
No open waterMoist soilWater
NBR
Land-cover and moisture changes; burn identification requires temporal comparison.
Ranges
Strong cover change / low moistureAltered cover / dry vegetationMore preserved vegetation cover
The entire cycle tracked index by index.
Every pass feeds time series per index. The combined reading shows the evolution of vigor, the change in area within each range, and the drop in moisture before the problem becomes visible in the field. The charts below illustrate the tracking format delivered on the dashboard and in reports.
Curva de vigor da safra
NDVI, NDRE e NDMI por data de passagem: emergência, pico vegetativo e senescência.
Distribuição de área por faixa de NDVI
Percentual da área em cada classe de vigor a cada passagem — mostra o fechamento do dossel.
Estresse hídrico acumulado
Evolução do NDMI: queda sustentada indica déficit hídrico antes de aparecer no vigor.
Composição atual da área
Última passagem processada: participação de cada faixa de vigor no total da área.
Assisted interpretation
A textual layer that makes the analysis actionable.
Language models organize and explain what the technical analysis has already quantified.
- 01Data and indices
- 02Organization
- 03Explanation
- 04Summary
- 05Report
- 01
Organizes findings
Detections and indicators are grouped and prioritized within the report.
- 02
Explains patterns
Index variations gain a description in operational language, with the caveats of each case.
- 03
Translates evidence
Numbers and maps become a summary that managers and technical teams understand.
- 04
Preserves distinctions
Observation, hypothesis and recommendation remain separated in the final text.
The language model supports the interpretation and communication of findings. The diagnosis still originates from the spectral analysis and computer vision.
Multisector applications
One vision layer, different operations.
Nexa Vision is a visual perception and interpretation layer able to work with different image sources. Spectral NDVI reading is one of the agricultural applications of this layer — not its definition.
SatelliteUAVFixed cameraOnboard cameraRGBNIRMultispectral
01
Agriculture
- Sentinel-2 and UAV
- RGB / NIR / multispectral
- NDVI, NDMI and red-edge
- Time series
- Stress, failures and anomalies
- Maps and reports
02
Food industry
- Visual monitoring of production steps
- Critical control points
- Hygiene and PPE use
- People and material flow
- Event logging and visual evidence
- Compliance-support reports
03
Mining
- Machine movement
- Routes and queues
- Cycle times
- Area utilization
- Congestion
- Shift and period comparison
04
Logistics
- Yard occupancy
- Docks and queues
- Cargo movement
- Dwell times
- Operational bottlenecks
05
Security
- Restricted zones
- Human presence
- PPE use
- Person/machine proximity
- Events and visual evidence
06
Livestock
- Counting
- Movement
- Spatial distribution
- Feed and water access
- Abnormal pattern detection
Nexa Vision observes; Nexa M1 interprets.
Conceptual flow: Sentinel-2 or UAV → Nexa Vision → anomaly or time series → Nexa M1 → interpretation → contextual diagnosis. Each product also operates independently.
Conceptual scheme · schematic reading, not to scale
Temporal dynamics
A single image shows a state. The series shows the history.
The same area is read across multiple passes; comparing them is what qualifies the diagnosis.
- T1
Vigor
Baseline reading for the cycle: initial distribution of vigor.
- T2
Moisture
The moisture index reacts to deficit before the eye can see it in the field.
- T3
Anomaly
A low-vigor patch stands out from the overall pattern of the area.
- T4
Trend
Comparison across passes confirms whether the anomaly advances or retreats.
- T5
Recovery
The priority zone returns to the area pattern after intervention.
T1T2T3T4T5
Operational delivery
From technical map to understandable report.
The delivery gathers what was observed, interpreted and recommended in a single document.
01
Maps and indices
02
Main findings
03
Period indicators
04
Priority zones
05
Technical conclusion
Nexa Vision01 / 01
Operational delivery
Visual analysis report
Cycle completeExecutive summary
Overview of the area for the period, with vigor and moisture variation per zone and comparison against the previous pass.
01
Main findings
02
Period indicators
Priority zonesTechnical conclusion
System outputs
What comes out of Nexa Vision.
Each analysis cycle delivers readings organized for different uses.
Visual diagnosis
A structured reading of the area, with maps and indices for the analyzed period.
Anomalies and zones
Spatial detections prioritized by severity and extent.
Temporal comparison
Evolution between passes, with trend and persistence per zone.
Structured report
A technical document with summary, findings, indicators and recommendation.
From drone to ground robot, in the same pipeline.
Nexa Vision integrates natively with Synapse Control and Axis, and also consumes images from third-party cameras, satellites and the technical team's mobile devices.
Inference where it makes sense for the operation.
Models run in the cloud for heavy analysis and at the edge, embedded in the robot, when response time matters for field navigation and actuation.
Nexa Vision applications in aerial and ground robotics
Aerial robotics
- —Autonomous inspection of fields and pastures with drones.
- —Obstacle and risk-area detection during the mission.
- —Herd counting and monitoring from the air.
- —Assessment of spray application and vegetative cover.
Ground robotics
- —Vision-assisted navigation along planting rows.
- —Pest identification and plant-level stand failure detection.
- —Infrastructure recognition: fences, water troughs, gates.
- —Targeted application and vision-guided sampling.
Computer vision for real operations.
The same vision and spectral-analysis technology adapts to different industries. Below, examples of how Nexa Vision is applied in each context.
Agriculture
Nexa Vision turns drone, satellite and field-camera imagery into precise agronomic diagnosis. Spectral indices detect water stress, nutritional deficiencies, pests, weeds, planting failures and yield variation within the field. The platform generates prescription maps for targeted application and tracks the crop's evolution across the cycle.
Livestock
Visual monitoring of the herd and pasture with drones and fixed cameras. Nexa Vision counts animals, estimates weight, identifies behavior patterns, detects welfare signals and maps forage availability and quality. Thermal imagery complements nighttime or low-visibility tracking.
Mining
Autonomous inspection of ore stockpiles, tailings dams, slopes, conveyor belts and infrastructure. Computer vision estimates volumes, identifies displacement, cracks, erosion and safety deviations, reducing human exposure and increasing inspection frequency in hard-to-reach areas.
Construction
Progress tracking, structural inspection, earthworks control and materials management. Nexa Vision compares site imagery with digital models and schedules to identify delays, inconsistencies and risks, and also monitors PPE use and the movement of people and machinery.
Infrastructure and Security
Autonomous patrolling, perimeter inspection and continuous monitoring of critical assets. Drones and ground robots capture images of fences, towers, lines, roads and facilities, while AI detects intrusions, damage, corrosion signs or obstructions and generates real-time alerts.
Logistics and Operations
Visual counting and tracking of loads, pallets, vehicles and equipment. Nexa Vision reads plates, identifies volumes, verifies positioning and integrates the data with management systems to speed up inventories, audits and yard control.
From image to report, a single chain of interpretation.