Journal of Experimental Agriculture International

Background: This research presents a significant stride in precision agriculture, focusing on the development and field evaluation of a self-propelled intra-row weeder engineered using mechatronics and machine learning. The study was motivated by the need for labor-efficient and environmentally friendly weed control methods, as conventional techniques pose various challenges.

Methodology: The intra-row weeder, equipped with a crop detection and avoidance system, was developed using a sensor, servo motor, encoder, weeding tool, and a microprocessor (Arduino Uno). A crop detection and avoidance algorithm, based on the K-nearest neighbor machine learning tool, was developed and trained using a customized feature method. This facilitated the system’s ability to accurately distinguish between plants and crops, a distinction that was programmed based on object height. This approach proved effective under the various conditions.

Results: Field performance evaluation of the weeder was conducted at different forward speeds and plant-to-plant spacing. The results revealed strong correlations between operating parameters and responses such as plant damage, weeding efficiency, performance index, and field efficiency, with R² values ranging from 67.87% to 83.61%. Optimal performance was achieved at a forward speed of 2.5 km∙h^-1 and plant spacing of 60 cm, yielding a field capacity of 0.041 ha.h^-1 and field efficiency of 86.25%. This study, therefore, provides a less labor-intensive solution for weed management in precision agriculture, paving the way for future innovations in the sector.