Read the crop, not only the conditions around it.
Scheduled cameras add the crop itself to RiceGuard's field record. Computer vision is designed to classify rice growth stage and locate pest-related patterns in field images.
Scheduled camera captures support growth-stage and pest-related analysis.
Status Method and intended outputs documented. Evaluation continues; no accuracy figure is published.
Diagram explains the method - it is not a live detection result.
The image remains traceable through every step.
- 01 Capture
Scheduled images from two 2 MP IP67 infrared cameras.
- 02 Timestamp
Keep the capture time and field source.
- 03 Compress
Reduce the image for local handling and transmission.
- 04 Pre-filter
Prepare the frame before model inference.
- 05 Analyse
Evaluate crop stage and pest-related patterns.
- 06 Synchronise
Send images, metadata and labels when connectivity is available.
Designed outputs Growth-stage class Predicted label · confidence score · highlighted image area
Seven stages turn an image into crop-cycle context.
- 01 Land preparation
- 02 Transplanting
- 03 Tillering
- 04 Stem elongation
- 05 Flowering
- 06 Grain filling
- 07 Maturation
Why stage matters Water, nutrient and pest evidence becomes more useful when read against the crop's observed stage—not only the calendar date.
Designed to identify and locate pest-related patterns.
A bounding box highlights the image area associated with a predicted pest-related label.
A confidence score shows how strongly the model supports that prediction; it is not guaranteed accuracy.
Records collected across Thai paddies.
- Brown planthopper counts per tiller
- Leaf blast percentage
- Sheath blight
- Bacterial leaf streak
Evidence boundary The dataset is used to develop and evaluate the model. Accuracy will be published only after formal evaluation is complete.
The labelled image set is uneven—and evaluation must reflect it.
- More images do not mean better accuracy. Tillering dominates the labelled set.
- Assess each stage separately. Flowering has a smaller evidence base.
- Accuracy remains unpublished. Formal evaluation is continuing.
Two-pole camera images support an experimental crop-stage view.
- 01 Input Two poles
Scheduled field images.
- 02 Context Image + context
Time and field source.
- 03 Interpretation Calculated stage
Development derived from images.
- 04 Output Phenology dashboard
Experimental system view.
Current status Operating experimental view · reading tower-camera images from two poles.
What is documented, operating and still under evaluation.
- Documented Image collection Scheduled capture · two 2 MP IP67 infrared cameras · timestamping · compression
- Documented Growth-stage design Seven rice stages and the role of crop-stage context
- Documented Pest evidence 1,214 records · 50 provinces · 30+ indicators
- Operating · experimental Phenology output Camera-derived growth stage from two poles
- Under evaluation Model accuracy No accuracy figure has yet been published
Images and sensor packets leave the field differently.
Continue to the two-radio strategy, local buffering and cloud architecture used for storage and fusion.