This research project focuses on grape segmentation using deep learning techniques in real-world vineyard conditions. The aim is to improve agricultural productivity by enabling accurate detection and segmentation of grape clusters.
The study presents a comparative analysis of YOLOv11, YOLOv12, and YOLOv26 to evaluate their performance in terms of accuracy and speed.
-
Laxmi Narayan | UniversityID 2211981212
-
Kumud Sharma | UniversityID 2211981207
-
University: Chitkara University
Research Project
- Perform accurate grape segmentation
- Compare YOLOv11, YOLOv12, and YOLOv26 models
- Evaluate performance using standard metrics
- Identify the best model for real-world agricultural use
- Python
- YOLO (Ultralytics)
- Deep Learning
- Computer Vision
- GPU Training
- Used a labeled dataset of vineyard images
- Applied instance segmentation techniques
- Trained all models using the same configuration for fair comparison
- Evaluated models based on accuracy and inference speed
- Precision
- Recall
- F1-Score
- mAP@0.5
- mAP@0.5:0.95
- Inference Time
- YOLOv11 → Highest accuracy
- YOLOv26 → Best balance between speed and performance
- YOLOv12 → Lower performance due to lack of pretrained weights
| Model | Precision | Recall | F1 | mAP@0.5 | mAP@0.5:0.95 | Inference (ms/img) | Epochs |
|---|---|---|---|---|---|---|---|
| YOLOv11 | 0.599 | 0.481 | 0.534 | 0.525 | 0.284 | 21.05 | 500 |
| YOLOv12 | 0.230 | 0.399 | 0.292 | 0.303 | 0.137 | 20.62 | 500 |
| YOLOv26 | 0.478 | 0.513 | 0.495 | 0.411 | 0.227 | 16.81 | 500 |
YOLOv11
YOLOv12
YOLOv26
- Model training completed
- Evaluation completed
- Research paper finalized
- Use larger and more diverse datasets
- Optimize models for real-time deployment
- Integrate with smart farming systems (drones/robots)
- YOLO Research Papers
- Deep Learning in Agriculture
- Computer Vision for Fruit Detection
This project is developed for academic and research purposes in the domain of precision agriculture.