MakindeAyomide/unified-vit-west-african-crops

Unified Multi-Crop Vision Transformer for West African Agricultural Disease Diagnosis # Unified Multi-Crop Vision Transformer for West African Agricultural Disease Diagnosis A Vision Transformer approach to cross-crop plant disease classification across six West African staple crops — developed for ICT 324, Bells University of Technology. ## Project Overview Existing plant disease detection models are overwhelmingly single-crop specialists, despite West African smallholder farms typically cultivating multiple crops together. This project builds a **single Vision Transformer** capable of diagnosing disease across **six economically important crops** — Cassava, Maize, Taro (substituting Yam), Tomato, Pepper, and Potato — rather than requiring six separate specialist models. **Final Results:** 81.21% test accuracy (Custom ViT) vs. 58.97% (Classical ML baseline) — a 22.24 percentage-point improvement. ## Team | Role | Member | |---|---| | Team Lead & Implementation Lead | Makinde Ayomide Daniel (2023/12127) | | Literature Reviewer | Sanni Samiat Ajoke | | Writing Coordinator | Sodamade Ismail | | Visualisation Specialist | Nweze Sharon | | Documentation Support | Shittu Qais Ayomide | Department of Computer Science, Bells University of Technology, Ota. ## Repository Structure ``` ├── notebooks/ │ └── ML_DL_Universal_Crop_Baseline.ipynb # Full pipeline: data prep, classical baseline, ViT training, evaluation ├── docs/ │ ├── Full_Manuscript.docx # Complete report: Intro, Lit Review, Methodology, Results, Discussion, Conclusion, References │ ├── ICT324_Project_Summary.docx # Condensed project summary │ ├── EDA_Report_UniPlant_WA.docx # Exploratory data analysis report │ ├── Results_Section.docx # Standalone Results section with per-class breakdown │ └── Verified_Reference_List (1).docx # 50-paper literature list, DOI-verified ├── results/ │ └── test_evaluation_report.txt # Final test-set classification report (all 29 classes) └── README.md ``` ## Dataset 67,443 …