PlantVision cv001dd

Overview

PlantVision is an innovative deep learning solution that leverages computer vision to help farmers and agricultural specialists identify plant diseases quickly and accurately. Built on a Convolutional Neural Network architecture, this model analyzes images of plant leaves to detect and classify diseases, providing early warning and treatment recommendations to prevent crop loss.

Technical Documentation

Architecture

The system consists of three main components:

CNN-based Image Classification Model: Built with TensorFlow and PyTorch, using an EfficientNet-B2 backbone pre-trained on ImageNet.
Mobile Application: Developed using Flutter for cross-platform deployment.
Backend API: A Flask-based API to handle image processing, model inference, and deliver results.

The CNN Model

The core of the project is the deep learning model. We chose EfficientNet due to its excellent balance of accuracy and computational efficiency, making it suitable for potential mobile deployment. This is a key architectural decision.

Data Preprocessing

Data augmentation was critical. Techniques included random rotations, flips, and brightness adjustments to simulate real-world conditions.

Model Performance

Below is a summary of the model's performance on the hold-out test set.

Metric	Score	Notes
Accuracy	98.7%	Overall correctness
Precision	97.2%	Of positive predictions, how many are correct
Recall	96.5%	Of actual positives, how many were found
F1-Score	96.8%	Harmonic mean of Precision and Recall

Note: These metrics represent the weighted average across all disease classes, accounting for class imbalance.

Mathematical Formulation

The loss function used is the cross-entropy loss, defined as:

L_{CE} = - \sum_{i=1}^{C} y_i \log(\hat{y}_i)

Where $C$ is the number of classes, $y_i$ is the true label (one-hot encoded), and $\hat{y}_i$ is the predicted probability for class $i$ .

Performance Metrics

Accuracy

98.7%

Precision

97.2%

Recall

96.5%

F1 Score

96.8%

Top 1% Acc

99.1%

Top 5% Acc

99.8%

Project Structure

Project Root

plantvision/
├── data/                    # ⛔ Not included in Git. Use DVC or git-lfs
├── src/                     # ✅ Source code
│   ├── configs/             # 📄 YAML configs (no hardcoded values)
│   ├── data/                # 📦 Data loading, augmentations
│   ├── models/              # 🧠 Model definitions: CNNs, EfficientNet
│   ├── train.py             # 🚂 Launch training with config + MLflow
│   ├── evaluate.py          # 📊 Validation + metrics
│   ├── quantize.py          # 🔧 Convert to ONNX/TFLite/INT8
│   ├── serve/               # 🌐 FastAPI or TorchServe entrypoint
│   └── utils.py             # 🛠 Logging, reproducibility, etc.
├── docker/                  # 🐳 Training + inference Dockerfiles
├── mlruns/                  # 🧪 MLflow experiment tracking
├── mlflow/                  # ⚙️ Config for MLflow server (optional)
├── Dockerfile               # 🔧 Main container spec
├── docker-compose.yml       # 🚀 Stack for local dev
├── requirements.txt         # 📦 Python dependencies
├── README.md                # 📘 Docs & instructions
├── start.sh                 # 🏁 Easy entrypoint to run things
└── experiment/              # 📁 Jupyter notebooks, exploration