Ironside35/Hurma_VGG16_Model.keras

🌴 VGG16 Transfer Learning Date Fruit Classifier

This is the 8th and final application in my Deep Learning series, exploring the limits of Transfer Learning using the massive VGG16 architecture on an agricultural dataset.

🧠 Model Strategy

• —Backbone: VGG16 (Pre-trained on ImageNet).
• —Strategy: Frozen convolutional base (Feature Extraction) combined with a newly trained Custom Dense Head for the 9 date fruit classes.
• —Classes: 9 distinct varieties (Ajwa, Galaxy, Medjool, Meneifi, Nabtat Ali, Rutab, Shaishe, Sokari, Sugaey).

📊 Key Metrics

• —Training Accuracy: %83.38
• —Comparison: Evaluated against a Custom CNN baseline to measure Transfer Learning efficacy on intra-class similarity.

🎯 Conclusion

While VGG16 is a powerhouse for general objects, it achieved 83.38% here, showing that its ImageNet-trained weights (optimized for cars, dogs, etc.) struggled slightly with the subtle, texture-heavy differences of date fruits compared to a scratch-built CNN.