MediDeepCF: A Multi-Task Deep Learning and Fuzzy Logic Framework for Maize Leaf Disease Detection and Severity Estimation
MediDeepCF is a multi-task deep learning framework developed for robust maize leaf disease detection and severity estimation. The pipeline integrates:
- Semantic segmentation using DeepLabV3+ with ResNet-50
- Disease classification using EfficientNet-B0 with CBAM attention
- Severity quantification using a fuzzy logic-based inference system
This work was conducted as part of an academic research project and achieves a high average F1-score of 96.51%, demonstrating strong performance and interpretability under real-world conditions.
- RGB median filtering for image denoising
- Multi-task architecture combining segmentation and classification
- Attention enhancement with CBAM
- Fuzzy rule-based disease severity estimation
- Stratified 4-fold cross-validation with detailed metrics tracking
- Publication-ready visualizations and performance graphs
- Python 3.10
- PyTorch, torchvision
- OpenCV, NumPy, Matplotlib
- EfficientNet, CBAM
- Scikit-learn, Scikit-fuzzy