Applied AI · Applied AI
TomatoVision
A computer-vision study that detects three greenhouse tomato maturity stages and compares a YOLOv11 baseline with modified models and a three-model WBF ensemble.
YOLOv11 Swin Transformer PyTorch OpenCV Weighted Boxes Fusion
The Problem
Greenhouse tomatoes overlap, hide behind leaves, and appear at different scales, making maturity detection difficult for a single detector configuration.
The Solution
Adjie evaluated a YOLOv11 baseline, Swin Transformer and multi-scale SPPF variants, then fused their predictions with Weighted Boxes Fusion while reporting single-model and ensemble results separately.
Key Capabilities & Workflow
- Detects green, orange, and red maturity classes in greenhouse imagery.
- Uses the YOLOv11 result as the evaluation baseline.
- Tests Swin Transformer and multi-scale SPPF architectural variants.
- Combines three model outputs with Weighted Boxes Fusion.
Verified Evidence
YOLOv11 baseline
0.795 [email protected]
Best modified model
0.807 [email protected]
Three-model WBF
0.824 [email protected]
WBF stricter metric
0.499 [email protected]:0.95
Why It Matters
The study shows which gains come from a single architecture and which require ensemble inference, rather than merging them into one headline result.
Confidentiality & Limitations: The portfolio reports evaluated research results and public project media; it does not claim a deployed agricultural product.