Adaptive Weed Classification in Agriclture: Dynamic Green Prior for improved Plant Localization and Accuracy.

Abstract

Adaptive techniques for weed classification in agricultural environments are necessary because they must take into account differences between crop appearance and variables related to the environment. This paper proposes a plant localization method based on dynamic green prior that is specially designed for weed classification applications. Unlike static methods, the system dynamically modifies the green detection parameters in response to seasonal variations, lighting conditions and real-time environmental cues. Using comprehensive testing on real-world agricultural datasets, the effectiveness of the dynamic method has been exhibited in precisely localizing plants and classifying weeds. The accuracy rate of the model trained using the preprocessed image is 99.27%. There is a significant improvement of 2% with the inclusion of pre-processing.