6

Multi-Plant Leaf Disease Detection & Segmentation

Multi-Plant Leaf Disease Detection & Segmentation

This project comprises a comprehensive AI workflow built to detect and segment a diverse array of leaf diseases across multiple plant species. It tackles two critical tasks: first, classifying up to 67 different disease categories using a DenseNet-201 architecture; second, pinpointing disease regions via semantic segmentation using a DeepLabV3+ model with ResNet-101 backbone. By combining both classification and segmentation in a unified pipeline, the system dramatically enhances disease identification precision and interpretability, providing detailed localization, not just a disease label.

The pipeline starts by ingesting curated leaf images from multiple Kaggle sources, organized into training, validation, and test sets. For disease detection, the project implements DenseNet-201 with robust data augmentation and custom training callbacks, achieving an impressive 97.03% classification accuracy across 67 disease classes. Simultaneously, the segmentation module leverages DeepLabV3+ enhanced with Atrous Spatial Pyramid Pooling (ASPP) and ResNet-101, trained on annotated leaf masks. Results include clear overlay visualization of disease areas, making it easier for researchers and practitioners to assess not just the presence of disease but its precise location and extent.

This project is valuable for both research and agricultural deployment. Its modular architecture is data-driven and extensible. New disease classes or plant types can be supported by retraining or fine-tuning on additional datasets. The inclusion of both classification and segmentation makes it a powerful tool for field diagnostics and precision agriculture, enabling early detection, severity assessment, and targeted treatment guidance. Moreover, the use of notebooks and clear documentation makes the pipeline accessible and reproducible—researchers, agronomists, or developers can easily explore, adapt, and build on this foundation.

Faculty

Students

  • Isra Mansoor
  • Muhammad Abdullah
  • Usama Athar

Comments are closed.