btr2006/Multi-Crop-Disease-Severity-Yield-Risk

AI-Based Multi-Crop Disease Severity & Yield-Risk Prediction System

An end-to-end Machine Learning project featuring three linked prediction models feeding a rule-based Yield-Risk Fusion Engine, wrapped in an interactive single-page Streamlit Application.

---

🌟 Features & System Architecture

                               ┌───────────────────────────────────┐
                               │       User Uploads Leaf Image     │
                               └─────────────────┬─────────────────┘
                                                 │
                        ┌────────────────────────┴────────────────────────┐
                        ▼                                                 ▼
        ┌───────────────────────────────┐               ┌───────────────────────────────────┐
        │  Module A — Disease Detection │               │   Module B — Severity Estimation  │
        │  MobileNetV2 Transfer CNN     │               │   HSV Color & Lesion Area Proxy   │
        │  (38 Crop & Disease Classes)  │               │   (Healthy / Mild / Mod / Severe) │
        └───────────────┬───────────────┘               └─────────────────┬─────────────────┘
                        │                                                 │
                        └────────────────────────┬────────────────────────┘
                                                 │
┌───────────────────────────────────┐            │
│  User Soil & Climate Inputs       │            │
│  (Temp, N, P, K, Fertilizer)      │            │
└─────────────────┬─────────────────┘            │
                  │                              │
                  ▼                              │
┌───────────────────────────────────┐            │
│   Module C — Yield Prediction     │            │
│   Random Forest / XGBoost Regr.   │            │
│   (Expected Yield in tonnes/ha)   │            │
└─────────────────┬─────────────────┘            │
                  │                              │
                  └──────────────────────┬───────┘
                                         │
                                         ▼
                       ┌───────────────────────────────────┐
                       │  Module D — Yield-Risk Fusion     │
                       │  Rule-Based Matrix & Recommender  │
                       └─────────────────┬─────────────────┘
                                         │
                                         ▼
                       ┌───────────────────────────────────┐
                       │   Final Streamlit Report Card     │
                       │   (6 Outputs + Lesion Overlay)    │
                       └───────────────────────────────────┘

1. Module A — Disease Detection Model: Pretrained MobileNetV2 transfer learning CNN fine-tuned on 38 plant disease classes (>90% accuracy).
2. Module B — Severity Estimation Engine: Standalone estimate_severity algorithm using HSV color segmentation and lesion area thresholding.
3. Module C — Crop Yield Regressor: Random Forest Regressor trained on soil nutrients ($N, P, K$), temperature, and fertilizer ($R^2 = 0.9908, \text{RMSE} = 0.1864$).
4. Module D — Yield-Risk Fusion Engine: Explainable rule matrix combining disease diagnosis, severity percentage, and expected yield vs historical benchmark ($8.53\text{ t/ha}$) to generate a 5-tier Risk Badge and 1-line actionable recommendation.
5. Streamlit Interactive UI: Single-page web dashboard displaying side-by-side leaf image + segmented lesion mask alongside 6 live result cards.

---

📁 Directory Structure

plant disese detection and crop yeld prediction/
├── data/
│   ├── disease/        # 38 Crop-Disease class folders (train/valid)
│   ├── severity/       # Plant Pathology 2021 competition dataset
│   └── yield/          # Crop Yiled with Soil and Weather.csv
├── models/
│   ├── disease_model.h5    # MobileNetV2 Keras CNN Model
│   ├── disease_labels.json # Index to Disease Class Mapping
│   ├── yield_model.pkl     # Fitted Random Forest Regressor
│   ├── yield_scaler.pkl    # StandardScaler for Soil Features
│   └── yield_stats.pkl     # Benchmark Yield Statistics
├── src/
│   ├── __init__.py
│   ├── disease_detection.py # Module A training & inference
│   ├── severity_estimation.py # Module B HSV lesion segmentation
│   ├── yield_prediction.py # Module C regression training & evaluation
│   └── risk_engine.py      # Module D yield-risk rule fusion engine
├── app/
│   └── main.py             # Streamlit Application Dashboard
├── tests/
│   └── test_severity.py    # Unit tests for severity estimation proxy
├── requirements.txt        # Python package dependencies
└── README.md               # User guide & execution instructions

---

🚀 Setup & Execution Instructions

1. Install Dependencies

pip install -r requirements.txt

2. Run Module Unit Tests

python -m pytest tests/test_severity.py -s

3. Launch Streamlit Application

streamlit run app/main.py