Aiming at the low detection rate,missed detections,and false detections in weed detection,an improved weed detection algorithm based on YOLOv8 is proposed.Add variable convolution to the backbone network and intro-duce learnable convolution kernels to effectively improve the recognition rate of weeds;Introducing the GCNet global attention mechanism,the model focuses more attention on the weeds to be detected,reducing missed and false detec-tion rates.Constructing a weed dataset and conducting experimental verification,the results showed that the improved algorithm improved accuracy and recall by 2.7％、2.2％and 5.5％respectively.It can effectively solve the problems of missed and false detections in weed detection,achieve precise weed detection,and provide reference for the de-velopment of intelligent weed control robots.
万方数据
维普
