【摘要】 针对海棠果加工依赖人工上料的问题，提出一种基于深度学习的海棠果定向方法，通过分析海棠果的物理特性与自然姿态概率，设计了一套由振动输送、视觉检测与气动分选组成的定向上料系统。首先利用振动实现果实初步姿态调整，接着视觉检测模型实时识别海棠果的姿态，气动分选则依据识别结果进行吹除回流。实验结果表明，在优化的振动参数下，系统整体定向成功率达90.4%,能够稳定地完成海棠果的定向输送，为后续自动化加工提供了可靠的技术支持。更多还原

【Abstract】 This study proposes a deep learning-based orientation method for Haitang fruit(Malus prunifolia).By analyzing the physical characteristics and natural postural distribution of the fruit, an oriented feeding system integrating vibration conveying, visual inspection, and pneumatic sorting was developed.Vibration was first applied to achieve preliminary posture adjustment of the fruit.A visual detection model then identified the fruit posture in real time, and pneumatic sorting was subsequently employed to remove and recycle misoriented fruits based on the detection results.Experimental results indicate that, under optimized vibration parameters, the system achieved an overall orientation success rate of 90.4%.The system consistently performed oriented conveying of Haitang fruit, thereby providing reliable technical support for subsequent automated processing.更多还原