Development and Evaluation of a Yolo Algorithm-Based Robotic Sprayer for Real-Time Weed Detection

Abstract Weed control with chemicals is a challenging process that should be performed in a rational way to reduce their negative impact on the surrounding environment. The growth of artificial intelligence algorithms encourages researchers to develop smart spraying robots that detect and spray weeds and distinguish them from the main crop which leads to sustainable use of these chemicals and achieves some of the sustainable development goals. However, few studies are available to comprehensively compare different versions of YOLO algorithm to detect weed. In this research, seven versions of YOLO algorithms were evaluated for their performance to detect and spray four types of weeds, namely, Cultivated licorice ( Glycyrrhiza glabra L.), Dyer’s Croton ( Chrozophora verbascifolia ), Lambsquarters ( Chenopodium album L.), and Puncturevine ( Tribulus terrestris L.) using a locally manufactured remotely controlled spraying robot. The results showed that YOLOv6n surpassed other algorithms which achieved the highest precision (0.89), recall (0.80), F1-score (0.84), [email protected] (0.86), inference speed (18.83 fps), in addition to the field indicators including true positive rate (0.83), false negative rate (0.17), false positive rate (0.19), true negative rate (0.81).