Aiming at the problems of low apple recognition rate and inaccurate localization in the current complex orchard environment, an apple target detection algorithm based on improved YOLOv8 was proposed in this paper to implement the recognition and localization of binocular vision system. Firstly, MobileNetV3 lightweight network is introduced into the backbone network, which improves the ability of the model to extract apple features. Secondly, the neck network introduces global attention mechanism to improve the detection accuracy of small target objects. Finally, the Focal-EIOUloss function is introduced to improve the problem of positive and negative sample imbalance. Experimental results show that the improved algorithm can effectively recognize and locate apples in complex environment.