Exploring YOLOv8 architecture applications for weed detection in crops

Abstract

This work has a goal to test a deep-learning approach to the problem of aerial weed detection in crops. The issue of this type of detection lies in the nature of plants and their life cycles. Crops as well as weeds change their appearance and can be similar in physical appearance. The use of advanced models like the You Only Look Once v8 (YOLOv8) allows for fast and accurate predictions. In this work, five different sizes of the YOLOv8 are applied to the same dataset consisting of aerial images of plants. The results, metrics, and actual predictions are provided for every of the five models. The modernization of the agricultural domain has begun, and the use of artificial intelligence (AI) is paramount to stay ahead of the competition. The experimental outcomes indicate significant potential of YOLO networks in this domain, and further possibility to integrate these networks with precision agriculture