Leaffliction - Plant Disease Classification

A computer vision project focused on classifying plant diseases through leaf image analysis, developed as part of the 42 School curriculum.

Project Overview

Leaffliction is a comprehensive machine learning pipeline that analyzes plant leaf images to identify various diseases. The project encompasses data analysis, augmentation, image transformation, and classification using deep learning techniques.

Key Features

1. Data Analysis & Visualization

• Automated dataset exploration and statistical analysis
• Generation of pie charts and bar charts showing class distribution
• Hierarchical directory structure analysis
• Identification of class imbalance issues

2. Data Augmentation

• Implementation of 6+ augmentation techniques:
  • Flip (horizontal/vertical)
  • Rotation
  • Skew
  • Shear
  • Crop
  • Distortion
• Automatic dataset balancing
• Preservation of original file naming conventions

3. Image Transformation Pipeline

• Advanced image processing using PlantCV or similar libraries
• Multiple transformation techniques:
  • Gaussian blur
  • Masking
  • ROI object detection
  • Object analysis
  • Pseudolandmark detection
  • Color histogram analysis
• Batch processing capabilities
• Flexible CLI with custom arguments

4. Classification System

• Deep learning model training for disease recognition
• Separation of training and validation datasets
• Model persistence and deployment
• Real-time prediction with visual output
• Target accuracy: >90% on validation set (minimum 100 images)

Technical Challenges

Data Management

• Handling imbalanced datasets across multiple plant species and disease types
• Managing large-scale image augmentation without quality loss
• Ensuring proper train/validation split to prevent overfitting

Image Processing

• Implementing robust leaf segmentation algorithms
• Handling varying image quality, lighting conditions, and backgrounds
• Extracting meaningful features from diverse leaf morphologies

Model Performance

• Achieving >90% accuracy requirement on validation data
• Preventing overfitting while maintaining generalization
• Optimizing model architecture for multi-class classification

Infrastructure

• Working within 42's cluster environment constraints
• Managing storage limitations (using goinfre when necessary)
• Creating SHA1 signatures for dataset integrity verification

Skills Developed

Computer Vision

• Image preprocessing and enhancement techniques
• Feature extraction from biological samples
• Understanding of color spaces and their applications
• Object detection and segmentation

Machine Learning

• Dataset preparation and curation
• Data augmentation strategies
• Model architecture selection and tuning
• Training pipeline development
• Overfitting prevention techniques
• Model evaluation and validation

Software Engineering

• Clean code architecture following coding standards (flake8 for Python)
• CLI design and argument parsing
• Batch processing systems
• File I/O operations and directory management
• Version control with Git
• Documentation and reproducibility

Domain Knowledge

• Plant pathology basics
• Disease classification categories
• Visual characteristics of plant diseases
• Agricultural computer vision applications

Project Structure

leaffliction/
├── Distribution.py         # Dataset analysis and visualization
├── Balance.py              # Dataset balancing utility
├── Augmentation.py         # Image augmentation tool
├── Transformation.py       # Image transformation pipeline
├── train.py               # Model training script
├── predict.py             # Inference and prediction
├── evaluate.py            # Model accuracy evaluation
├── model_loader.py        # Automatic model ZIP extraction
├── model_output.zip       # Trained model (compressed)
├── model_output.zip.sha256 # Model integrity checksum

Requirements

• Python 3.x (recommended) or language of choice
• Machine learning libraries (TensorFlow/PyTorch/Keras)
• Image processing libraries (OpenCV, PIL, PlantCV)
• Visualization libraries (Matplotlib, Seaborn)
• Code must follow flake8 standards if using Python

Usage Examples

# Analyze dataset distribution
./Distribution.[extension] ./Apple

# Augment a single image
./Augmentation.[extension] ./Apple/apple_healthy/image.JPG

# Transform images in batch
./Transformation.[extension] -src Apple/apple_healthy/ -dst output/ -mask

# Train the model
./train.[extension] ./Apple/

# Make predictions
./predict.[extension] ./Apple/apple_healthy/image.JPG

Validation & Submission

• All code must be submitted via Git repository
• Dataset must NOT be included in the repository
• signature.txt file containing SHA1 hash of dataset.zip is mandatory
• Signature verification during evaluation (mismatches result in grade 0)
• Model must achieve >90% accuracy on validation set

Learning Outcomes

This project provides hands-on experience with the complete machine learning workflow, from raw data to deployed model. It emphasizes the importance of data quality, proper validation techniques, and the practical challenges of real-world computer vision applications in agriculture and plant pathology.

---

Developed as part of the 42 School curriculum - A project that transforms understanding of both computer vision and plant disease diagnostics.