Hybrid Vision Transformer and CNN-Based System for Real-Time Weed Detection in Precision Agriculture

Weeds are a serious problem for agriculture, as they result into lower yields, higher costs, and excessive usage of herbicides which are destructive to land and to people respectively. This work proposes a new approach in weed detection through the use modern machine learning approaches due to the aforementioned problems with a success rate of 98 percent. The system integrates three innovative algorithms: the Hybrid Vision Transformer-CNN Model adopted for selective feature extraction, the Adaptive Multi-Spectral Feature Fusion (AMFF) that enhances feature fusion from spectral data to differentiate weeds from crops, and the Hierarchical Weed Classification Algorithm, which provides further weed categorization. Intended for farmers, agronomist, as well as developers of agri-food technology, this system enables efficient usage of herbicides, minimizing hand work, and improving agricultural output. This project is a feasible solution to world precision agriculture issues because it has the capacity to give real-time, large-scale detection of weeds.