🌱 Revolutionizing Cassava Disease Detection with Deep Learning
Cassava is a staple crop for millions, especially in sub-Saharan Africa, where it provides a crucial source of food and income. However, diseases such as Cassava Mosaic Disease (CMD) and Cassava Bacterial Blight (CBB) threaten yields, leading to severe food security challenges. Traditional disease detection relies on manual inspection, which is often inaccurate, time-consuming, and not scalable.
With advancements in Deep Learning, we can now automate disease detection using Convolutional Neural Networks (CNNs) trained on cassava leaf images. 🚀
🔍 The Aim of This Study
The goal was to develop and evaluate CNN models for detecting cassava leaf diseases using image-based classification. The research also compared different CNN architectures, optimizers, and loss functions to determine the best-performing model for accurate disease identification.
📊 Dataset & Methodology
A publicly available Kaggle dataset was used, consisting of over 20,000 cassava leaf images, labeled into:
- Healthy Leaves
- Cassava Mosaic Disease (CMD)
- Cassava Brown Streak Disease (CBSD)
- Cassava Green Mottle (CGM)
- Cassava Bacterial Blight (CBB)
Key Steps in Model Development
1. Data Preprocessing & Augmentation
- Images were resized to 227x227 pixels
- Normalization was applied to ensure consistent pixel values
- Augmentation techniques (rotation, zooming, flipping) helped combat class imbalance
2. CNN Architecture & Model Training
- Multiple CNN architectures were tested
- Adam optimizer & RMSprop were compared
- Loss functions: Categorical Cross-Entropy vs Focal Loss
- Hyperparameter tuning to improve accuracy
3. Evaluation Metrics
- Accuracy, Precision, Recall, and F1-score were used to evaluate model performance.
- MLflow was used for experiment tracking and parameter logging.
🚀 Results & Key Findings
✅ Best Model Achieved 93% Accuracy
- A CNN model using Adam optimizer with Categorical Cross-Entropy loss performed best.
- The model successfully classified diseases, but CMD was overrepresented, indicating the need for more balanced datasets.
✅ Adam Outperformed RMSprop
- Models trained with Adam converged faster and achieved better accuracy.
✅ Focal Loss Did Not Outperform Categorical Cross-Entropy
- Contrary to expectations, Focal Loss did not improve results, highlighting the importance of dataset balancing techniques.
✅ Overfitting Remained a Challenge
- Despite using dropout layers and data augmentation, the model tended to favor CMD due to dataset imbalance.
🔬 Future Work & Improvements
- Hybrid Approaches: Combining CNN with XGBoost or Attention Mechanisms.
- More Data: Expanding the dataset for better generalization.
- Severity Assessment: Extending the model to predict disease severity for better agricultural interventions.
- Mobile App Deployment: Enabling real-time disease detection for farmers using smartphones.
🌍 Why This Matters
This research demonstrates the power of AI in agriculture, paving the way for scalable, automated, and cost-effective disease detection systems. By leveraging deep learning, farmers and agronomists can detect diseases early, improve yield, and combat food insecurity.