Skip to content

Latest commit

 

History

History
79 lines (60 loc) · 2.91 KB

MODEL.md

File metadata and controls

79 lines (60 loc) · 2.91 KB

MODEL.md - Egg Fertility Detection Model Documentation

Model Architecture

The egg fertility detection model is based on YOLOv8n, optimized for high-speed, single-column conveyor belt egg processing.

  • Base model: YOLOv8n
  • Task: Object Detection (Binary Classification: Fertile/Infertile)
  • Output: 2 classes (Fertile 'f', Infertile 'i')

Training Process

Dataset

  • Total samples: 179 images (139 training, 40 validation)
  • Features: Preprocessed egg images from single-column conveyor belt setup
  • Classes: 2 (Fertile, Infertile)

Training Configuration

  • Framework: PyTorch (via Ultralytics YOLO implementation)
  • Image size: 640x640
  • Batch size: 16
  • Maximum epochs: 300
  • Early stopping patience: 15
  • Learning rate strategy:
    • Initial learning rate (lr0): 0.01
    • Final learning rate factor (lrf): 0.001
  • Optimizer: SGD (Stochastic Gradient Descent)
  • Data augmentation:
    • Close mosaic: 10
    • Auto augment: randaugment
  • Mixed precision training: Enabled (amp=True)

Model Performance

Test Dataset

  • Number of test images: 20
  • Test environment: Single-column conveyor belt setup

Test Results

Based on the evaluation of the 20-image test dataset:

  • Accuracy: 1.00 (100%)
  • Precision: 1.00 (100%)
  • Recall: 1.00 (100%)
  • F1-score: 1.00 (100%)

These metrics indicate perfect performance on the test dataset, suggesting that the model has learned to classify egg fertility with high accuracy in the specific conveyor belt environment.

Speed Performance

  • Preprocess: 3.0ms per image
  • Inference: 10.0ms per image
  • Postprocess: 2.0ms per image
  • Total processing time: 15.0ms per image

The model demonstrates fast processing times, suitable for high-speed conveyor belt operations.

Multiple Detection Handling

The model occasionally detects a single egg as two objects. This is a known trade-off for achieving higher processing speed and lower computational costs. Important points:

  1. The conveyor belt processes eggs in a single column, ensuring only one egg is present in each image.
  2. Correction code is implemented in the testing scripts to handle multiple detections:
    • The highest confidence detection is used for classification.
    • Multiple detections are merged or filtered out.
  3. The system maintains high accuracy in egg type classification (fertile vs. infertile) despite occasional multiple detections.

This approach allows for:

  • Faster processing speeds
  • Lower computational requirements
  • Reduced AWS service costs
  • Maintenance of high classification accuracy

Future Work

  1. Expand the test dataset to include a larger and more diverse set of egg images
  2. Fine-tune the model to further reduce multiple detections while maintaining speed
  3. Conduct long-term performance monitoring in production environment
  4. Explore potential for model quantization to further optimize inference speed
  5. Investigate adaptive thresholding techniques to improve robustness across varying lighting conditions