The pole is designed to process, prioritise and remember.
RiceGuard processes field data at the pole before sending it onward. It checks readings, prepares images, stores recent records and can raise critical alerts even when connectivity is interrupted.
What the pole handles locally.
- 01 Reliable records Validated, timestamped and traceable
- 02 Offline operation Seven days stored; critical alerts local
- 03 Power-aware computing ESP32 always on; GPU work scheduled
Processed at the pole before it reaches the cloud.
- 01 Capture
Collect a scheduled sensor reading or field image.
- 02 Check
Validate values and flag anomalies.
- 03 Add context
Attach the time and source.
- 04 Prepare
Compress and pre-filter images.
- 05 Alert or store
Raise urgent alerts; store routine data.
Sensor readings and images are prepared differently.
- 01 Read Collect the configured values
- 02 Check Validate values and attach anomaly flags
- 03 Record Timestamp and store as a local time series
- 01 Capture Take the field image
- 02 Prepare Compress and pre-filter
- 03 Infer Pass the image to local AI
Shared rule Every output keeps its timestamp and source.
Always-on control. Image AI only when needed.
- Schedules sensor readings
- Manages power and station health
- Keeps local control running during link loss
- Runs TensorRT image inference
- Processes prepared camera frames
- Runs as a higher-power computing task
Power principle The ESP32 stays available; GPU work is scheduled only when needed.
A lost connection delays delivery - it does not erase the record.
Up to seven days of readings remain stored locally, and critical alerts can still be created.
- 01 Connected Normal sampling Readings are prepared and sent.
- 02 Link lost Local operation continues Sampling and health checks continue.
- 03 7-day buffer Recent history retained Routine records wait locally.
- 04 Link returns Records synchronise Stored history rejoins the wider platform.
The pole handles urgent information first.
- P1 · Critical Raise a local alert
A critical condition can be identified without waiting for the cloud.
- P2 · Presence / status Keep the station visible
Health and status show whether the pole is functioning.
- P3–P4 · Routine Store and queue
Normal readings and prepared images wait for transmission.
Two data types. Two model paths. One combined interpretation.
- Timestamp
- Source identity
- Crop context
Structured and visual features remain identifiable, but can now be read together by later system outputs.
Plain-language meaning The station first understands numbers and images separately. It combines them only after both are attached to the same field, time and crop context.
Where the local record goes next.
- 01 Transmit
LoRa carries small sensor packets; cellular carries images and bulk synchronisation.
- 02 Retain
Cloud services preserve the resulting field history.
- 03 Present
Agricultural context supports operator, farmer and phenology views.
Local preparation reduces unnecessary data, preserves traceability and allows important alerts to be raised even when field connectivity is interrupted.
Processing prepares the image. Vision interprets the crop.
Continue to the image pipeline, field dataset and documented computer-vision objectives.