Weed Detection using Vision Transformers
B.Tech thesis - ViT-backed DETR for weed detection, published at IEEE ASIANCON
🛠️ Tech Stack
- Languages: Python
- Concepts: Vision Transformers (ViT), Detection Transformer (DETR), Deep Learning, Object Detection
📌 Overview
Final year B.Tech thesis project formulating weed detection as two supervised learning tasks, classification and object localization/detection, for real-time smart farming applications.
✨ Key Highlights
- Proposed a Vision Transformer (ViT) as the feature extractor backbone for Detection Transformer (DETR), applying regularization techniques to mitigate overfitting.
- Achieved mAP@0.5 of 82.9% on the CottonWeed dataset, a 7% improvement over state-of-the-art CNN-based detectors.
- Published a survey paper at IEEE ASIANCON, providing a detailed taxonomy of weed detection techniques and the future potential of ViT in real-time smart farming solutions. [Paper]