Apple Detection in Complex Scenes Based on Improved YOLOv8

In order to enable the picking robot to quickly and accurately detect apple fruits of different sizes under complex environments in the orchard (e.g., scenarios with different lighting conditions, fruit leaf occlusion, and dense small targets at long distances), this paper proposes an improved YOLOv8-based algorithm. Firstly, the EMA attention module is integrated into the YOLOv8 network architecture, aiming to enhance the model's focus on the target fruit region, and at the same time, reduce the interference of non-critical features such as background and branch and leaf occlusion, so as to effectively improve the accuracy of fruit detection in the case of occlusion. Secondly, the fusion weighted bidirectional feature pyramid (BiFPN), through weighted feature fusion, enables the network to better learn the features of small targets and reduce the leakage detection rate of small targets. The experimental results prove that the improved algorithm outperforms the original model in most of the metrics, providing a reference for apple detection in complex scenes.