Z-YOLO: an apple object detection algorithm in orchard scenarios based on improved YOLOv11

In the agricultural production process of apple picking in complex orchard environments, there are often severe occlusion caused by the intertwined fruits and branches and leaves, as well as the small target detection problem in the far field of view due to the inconsistent growth distribution of fruits. In response to these issues, this study constructed an apple dataset in a complex orchard environment using an RGB color camera and proposed an apple detection network based on SimAm and an improved C3k2, named Z-YOLO. Based on YOLOv11, Z-YOLO incorporates a lightweight SimAm attention mechanism and proposes a C3k2_SC module to reduce redundant calculations and replaces the original C3k2 module, improving the model's accuracy while reasonably controlling the model parameters. Experiments show that Z-YOLO achieves an accuracy (P) of 85.7%, a recall rate (R) of 81.4%, and an average accuracy (mAP50) of 90.5% in the apple dataset of the complex orchard environment. Compared with the baseYOLOv11, P improves 1.4%, R improves 0.4%, and the average accuracy improves 1%. The provided model can effectively detect apples and has high robustness, providing technical support for the research and development of apple picking robots.