CHERRY-YOLO: LIGHTWEIGHT MODEL FOR CHERRY DETECTION BASED ON IMPROVED YOLOv8n

ABSTRACT At present, in the field of cherry recognition, problems such as dense fruit growth and frequent occlusions of branches and leaves affect the accuracy of the recognition process. To address these issues, this paper proposes a lightweight cherry detection model based on an improved YOLOv8n (CHERRY-YOLO). By introducing a FasterNet block module based on partial convolution (PConv) and replacing the bottleneck module in YOLOv8n's C2f with a FasterNet block module, a new C2f-Faster module is formed. As the module reduces the number of parameters and calculations, it also ensures high detection accuracy. By adding a convolutional block attention module (CBAM) between the SPPF module and C2f-Faster module in the backbone network area, the ability to extract target features is enhanced before multiscale feature map fusion, further improving the target detection accuracy. The experimental results show that CHERRY-YOLO has more advantages than the original YOLOv8n model. Its precision, [email protected], [email protected]:0.95, and recall increase by 1.3%, 1.0%, 0.6%, and 0.4%, respectively. Moreover, the number of parameters, computational size, and weight decrease by 21.27%, 20.99%, and 20.63%, respectively. The results demonstrate that CHERRY-YOLO provides important technical support for automatic cherry picking.