Weed Identification from Multispectral Agricultural Images

Weed control plays a critical role in agriculture, with implications for crop yields and farm incomes. The conventional processes such as manual weeding or blanket herbicide use are ineffective and environmentally unfriendly. This project involves a precision weed detection system based on multispectral imaging and machine learning. YOLO model and NDVI masks were utilized to detect and segment weeds from multispectral images. The NDVI masks made crops visible, which allowed YOLO to identify weeds with very high accuracy, attaining an average F1-score of 92%. Weed localization was also cross-validated against manual surveys with less than 5% error. This system provides a scalable, automated, and environmentally friendly solution for weed management.