Refined oil · smart supervision solution
Refined oilsmart supervision solution
For refined-oil circulation supervision, use existing gas-station video as an objective data entry: recognize vehicles, plates, arrival and departure, and nozzle raise/hang events; link them with transaction, payment, invoice, and declaration data to form a perceive–verify–alert–respond–feedback loop.
Real operations perception and supervision loop
Real fueling behavior · eventized perception
Real behavior perception
Vehicle, plate, nozzle state, and time
Event search and review
One event maps to one video clip
Rule-based alerts
Cross-source matching yields risk clues
Continuous sample iteration
Field issues feed algorithm improvement
Solution overview
DeepGlint’s refined-oil smart supervision solution is built on the visual intelligence workshop. It reuses existing gas-station cameras and networks; GBOX handles station-side video ingest and real-time inference, turning arrival, nozzle raise/hang, and departure video into standardized fueling events. EXPERT orchestrates devices, algorithms, rules, tasks, alerts, and response workflows. MENTOR handles sample feedback, annotation, training, evaluation, and model conversion. The solution gives regulators an independent, objective, reviewable data entry for compliance verification, risk discovery, work-order handling, and cross-department governance.
Application scenarios
Deliver supervision applications around fueling perception, operating-chain verification, risk assessment and response, and collaborative governance.
Intelligent fueling perception
Recognize vehicles, plates, arrival and departure, and nozzle raise/hang moments, turning continuous video into searchable, reviewable standardized fueling events.
Transaction and invoicing verification
Link transactions, payments, invoices, and filings by station, vehicle, nozzle, and time to surface missing links and timing anomalies for verification.
Risk clue assessment and response
From a clue, supervisors open the matching event and video clip to review, assign tasks, handle on site, collect remediation feedback, and archive results.
Cross-department collaborative governance
Vehicle trajectories, event records, and traffic data can be reused under role-based authorization to support tax, commerce, market regulation, and emergency management.
Solution advantages
Combine video eventization, low-code orchestration, cloud-edge collaboration, and a data closed loop to build continuously improving visual supervision for refined oil.
Reuse existing video assets
Reuse existing cameras, NVRs, and networks to reduce front-end retrofit and duplicate construction.
Real behavior as events
Turn continuous video into standard events for vehicles, nozzle state, and time so each visit can be searched and reviewed.
Multi-source verification
Link events with transaction, payment, invoice, and declaration data to create an independent, objective verification entry.
Low-code rule orchestration
Flexibly combine algorithms, frequency, timing, and response actions against regulation rules to adapt quickly to business change.
Lightweight cloud-edge deployment
GBOX runs real-time inference at the station with weak-network autonomy; the center pushes models and aggregates key events.
Continuous field-sample evolution
Low-confidence, false-positive, and miss samples flow back to MENTOR for annotation, training, evaluation, and redeployment.
Solution architecture
Built on the visual intelligence workshop stack to connect central training, algorithm operations, station inference, and supervision applications.
Business apps
- Overview
- Station profile
- Event search
- Risk alerts
- Work orders
- Analytics
EXPERT
Algorithm operations center / supervision hub
- Device management
- Algorithm management
- Task orchestration
- Rule engine
- Alert operations
- Evidence archive
MENTOR
Algorithm training center / continuous field evolution
- Sample feedback
- Data cleaning
- Assisted labeling
- Model training
- Evaluation
- Model conversion
GBOX
Station-edge intelligence
- Video ingest
- Real-time inference
- Weak-network autonomy
- Model push
- Event generation
- Clip upload
Station and business data
- Cameras / NVR
- Pumps / tank gauge
- POS / transactions
- Payment / invoice / filing
Data closed loop
Field event → central verification → risk response → sample feedback → model improvement → station redeployment
Events supply verification clues; final supervision decisions are made by staff reviewing video and operating records together.
Solution value
Move video assets from passive storage to proactive discovery, objective verification, and continuous operations.
From passive to proactive discovery
Continuously perceive real fueling behavior and automatically form verification clues for abnormal chains, reducing manual patrols and screen watching.
From experience to evidence-based verification
Jump from a clue to the matching event, video, and linked business records, with full traceability of handling and outcomes.
From siloed builds to collaborative scenarios
One capture produces standard event data that, within authorization, supports tax, commerce, market regulation, and emergency collaboration.
From one-off delivery to continuous algorithm ops
Field samples continuously feed training and evaluation so models evolve with station environments, business rules, and supervision needs.