.gif){kind=link}
.gif)
... Demonstration of real-time potato disease classification with 98.5% accuracy
This project implements a CNN-based solution for detecting potato diseases (Early Blight, Late Blight) with 98.5% accuracy. Developed as a practical implementation of my research on "Applying Machine Learning to Agriculture in Sri Lanka: Challenges and Outcomes", it addresses critical challenges in smallholder farming by providing real-time disease diagnostics.
- Accurate Disease Detection: CNN model with 98.5% validation accuracy
- Real-time Prediction: <3 second response time
- Farmer-Friendly Interface: Simple image upload and clear results
- Full-stack Architecture: From data processing to web deployment
- Optimized for Edge Devices: Model quantization for mobile use
Component Technology
Deep Learning TensorFlow, Keras
Backend API FastAPI, Python 3.9
Frontend React, Material-UI
Model Serving TensorFlow Serving, Docker
The model was trained on 3,152 potato leaf images from the PlantVillage Dataset with custom augmentations for Sri Lankan growing conditions:
Backend Setup
git clone https://github.com/piyuminadee/Potato-Disease-Classification.git
cd api
pip install -r requirements.txt
uvicorn main-tf-serving:app --reload
Frontend Setup
cd frontend
npm install --from-lock-json
npm run start
Model Serving with Docker
- docker build -t my-tf-serving
- docker run -it --rm
-v "D:\My-Code\MlProjects\potato_desease\models\potato_disease_savedmodel:/models/potato_model/1"-p 8606:8501tensorflow/serving:2.14.0--model_name=potato_model `--model_base_path=/models/potato_model
This project implements Section 2.2 ("Machine Learning in Agricultural Prediction/Detection") from my research paper: "Applying Machine Learning to Agriculture in Sri Lanka: Challenges and Outcomes" ResearchGate Link