Kibet-Rotich/plantdisease

# 🌱 AgriGuard Autonomous ML Service **An End-to-End Autonomous Foliage Pathology Classification & Retraining Pipeline** Live Demo Url --- ## 📖 Project Overview AgriGuard is an autonomous, production-grade computer vision service engineered to classify plant foliage diseases from non-tabular image data (PlantVillage dataset). The system transitions a static convolutional neural network into a resilient cloud architecture featuring: * **Real-time Single Prediction API:** Sub-second diagnosis across *Healthy*, *Early Blight*, and *Late Blight* Tomato classes. * **3-Biomarker Feature Storytelling Engine:** Automatically extracts and visualizes physical pathology ratios (HSV Chlorosis degradation, Canny edge structural fragmentation, and Otsu necrotic lesion surface area) to explain *why* the neural network made its decision. * **Autonomous Background Retraining:** An interactive UI control where users stage bulk zip dataset uploads and trigger asynchronous fine-tuning cycles without causing server downtime or API blocking. * **Horizontal Docker Scaling & Load Balancing:** Containerized with FastAPI, Uvicorn, and Nginx reverse-proxy load balancing to handle thousands of concurrent IoT field sensor requests. --- ## 🗂️ Repository Directory Structure Strictly structured according to production ML engineering standards: ```text plantdisease/ ├── README.md # Comprehensive project documentation & results ├── docker-compose.yml # Multi-container scaling & Nginx load balancer ├── locustfile.py # High-concurrency IoT stress simulation script ├── backend/ │ ├── Dockerfile # Keras 3 / FastAPI container blueprint │ ├── requirements.txt # Pinned Python ML dependencies │ ├── main.py # Asynchronous API routes & background task handlers │ ├── src/ │ │ ├── preprocessing.py # Byte decoding & 3-biomarker feature extraction │ │ ├── model.py # Keras 3 auto-discovery & fine-tuning engine │ │ └── pred …