Capture
- RGB images
- Multispectral NIR images
- Thermal images
- Acoustic samples
- Fruit and postharvest metadata
Technology
How AIDurian Works
A multimodal AI system that sees, hears, and understands durian.
From Capture to Action
Every AIDurian assessment follows the same four-stage pipeline.
Analyze
- Computer vision
- Acoustic spectrogram analysis
- Deep learning
- Spectral analysis
- Data preprocessing
Assess
- Maturity
- Ripeness
- Defects
- Shape
- Locule count
Act
- Harvest timing support
- Sorting and grading support
- Quality documentation
- Postharvest decisions
- Traceability workflows
Two Concepts, One System
Maturity and ripeness are different quality dimensions — AIDurian assesses both.
Physiological development
Maturity refers to the physiological development stage of the fruit — the point at which it can be harvested and will ripen properly afterward. It is relevant to harvest timing, export eligibility, and shelf-life potential.
Eating quality
Ripeness refers to the current eating-quality stage — texture, aroma, and flavor — and how ready the fruit is for immediate consumption. It is assessed through acoustic knock analysis of the fruit's internal condition.
Five AI Assessment Capabilities
Each module is trained on locally collected Puyat durian data and validated under the test conditions reported in the AIDurian Terminal Report.
Maturity
- Input
- Multispectral NIR
- Output
- Immature, mature, overmature
- Use
- Harvest timing and shelf-life potential
Ripeness
- Input
- Tapping sound
- Output
- Model class and confidence
- Use
- Current eating-quality assessment
Defect Detection
- Input
- RGB image or video
- Output
- Surface-defect classification
- Use
- External quality screening
Shape
- Input
- Full-fruit RGB image
- Output
- Regular or irregular morphology
- Use
- Grading and presentation assessment
Locule Analysis
- Input
- RGB top-view image
- Output
- Locule segmentation and count
- Use
- Internal structural assessment
Note: locule analysis uses a top view of the intact fruit — no opening required — and is non-destructive like the whole-fruit modules.
Multimodal Sensing
No single signal represents all aspects of durian quality — so AIDurian listens, looks, and measures.
RGB Imaging
Visible-light images capture surface appearance, shape, and defects — the foundation of visual assessment.
Multispectral NIR
Eight non-contiguous bands spanning 720–980 nm reveal physiological signals invisible to the eye.
Thermal Imaging
Surface temperature patterns support postharvest and condition assessment.
Acoustic Analysis
Standardized tapping recordings, converted to spectrograms, carry ripeness information through sound.
Metadata
Fruit origin, harvest round, and handling records give every assessment documented context.
Why multiple signals?
AIDurian combines multiple sensing methods because no single signal represents all aspects of durian quality. Together, RGB, multispectral NIR, thermal, and acoustic data give a fuller picture of each fruit.
See the Results →The Grading Station
The AIDurian Grading Station is a controlled imaging and assessment environment that standardizes fruit positioning, lighting, and capture for repeatable testing — currently a research prototype available for collaboration.
Current status and limitations: AIDurian is a validated research prototype. Reported accuracies were achieved under the specific test conditions described in the Terminal Report. Field deployment, scale-up, and standardization work continue through industry collaboration.