Strawberry Ripeness Classification Model

Model Description

This is a 4-class strawberry ripeness classification model trained on PyTorch with 91.71% validation accuracy. The model classifies strawberry crops into four ripeness categories:

unripe: Green, hard strawberries not ready for picking
partially-ripe: Pink/red, firm strawberries
ripe: Bright red, soft strawberries ready for picking
overripe: Dark red/brown, mushy strawberries past optimal ripeness

Training Details

Architecture: EfficientNet-B0 with custom classification head
Input Size: 128x128 RGB images
Training Epochs: 50 (early stopping at epoch 14)
Batch Size: 8
Optimizer: Adam with cosine annealing LR scheduler
Dataset: 2,436 total images (889 strawberry crops + 800 Kaggle overripe images)
Validation Accuracy: 91.71%
Training Time: ~14 epochs with early stopping

Usage

import torch
from torchvision import transforms
from PIL import Image

# Load model
model = torch.load("best_enhanced_classifier.pth")
model.eval()

# Preprocessing
transform = transforms.Compose([
    transforms.Resize((128, 128)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

# Classify image
image = Image.open("strawberry_crop.jpg")
input_tensor = transform(image).unsqueeze(0)

with torch.no_grad():
    output = model(input_tensor)
    predicted_class = torch.argmax(output, dim=1).item()

classes = ["unripe", "partially-ripe", "ripe", "overripe"]
print(f"Predicted ripeness: {classes[predicted_class]}")

Model Files

classification_model/best_enhanced_classifier.pth: Trained PyTorch model (4.7MB)
classification_model/training_summary.md: Detailed training metrics and results
classification_model/enhanced_training_curves.png: Training/validation curves

Integration

This model is designed to work with the strawberry detection model for a complete picking system:

Detection: YOLOv8 finds strawberries in images
Classification: This model determines ripeness of each detected strawberry
Decision: Only pick ripe strawberries (avoid unripe, partially-ripe, and overripe)

Performance Metrics

Class	Precision	Recall	F1-Score
unripe	0.92	0.89	0.91
partially-ripe	0.88	0.91	0.89
ripe	0.94	0.93	0.93
overripe	0.96	0.95	0.95

Overall Accuracy: 91.71%

Dataset

Source: Mixed dataset with manual annotations + Kaggle fruit ripeness dataset
Classes: 4 ripeness categories
Total Images: 2,436 (train: 1,436, val: 422)
Preprocessing: Cropped strawberry regions from detection model

Requirements

PyTorch >= 1.8.0
torchvision >= 0.9.0
Pillow >= 8.0.0
numpy >= 1.21.0

License

MIT License - see main repository for details.

Contact

For questions or improvements, please open an issue in the main repository.