π Sharing my end-to-end ML project: Krishi Rakshak β a crop disease classifier built with efficientnet_b0 for smallholder farmers.
βοΈ Model Highlights:
90%+ accuracy on PlantVillage (38 classes)
Trained with AdamW, scheduler, early stopping
Used weighted loss to handle class imbalance
π Evaluation:
Manual metric validation + confusion matrix
Precision, recall, F1, per-class accuracy
π§© Deployment:
PyTorch + Gradio UI, multilingual support
Designed for light, modular inference
π GitHub: https://github.com/VIKAS9793/KrishiRakshak.git
πΉ Demo: https://drive.google.com/file/d/1PDxYq5rOuCXZAldZlSd6Q3M5WIGXEtbJ/view?usp=drivesdk
Built independently from scratch β feedback welcome from ML/AgriTech community on optimization or scaling for noisy field data.
Top comments (0)