With the continuous development of deep learning and computer vision technologies, the integration of image recognition with robotic manipulation has become an important topic in smart manufacturing and agricultural automation. In this study, we propose an interactive teaching platform for fruit classification and autonomous grasping, which incorporates a vision sensor (camera) with the You Only Look Once Version 7 (YOLOv7) object detection algorithm enhanced by the adaptively spatial feature fusion (ASFF) module. The camera serves as the primary sensing component, providing a real-time visual input that is processed by the detection algorithm to enable robust fruit recognition and precise grasping actions. The system integrates advanced visual recognition, adaptive grasping strategies, and a user-friendly human–machine interface, creating a practical learning environment for hands-on experience in intelligent system applications. Experimental results indicate that the proposed YOLOv7-ASFF model, trained and validated on a self-constructed fruit image dataset, achieved a mean average precision (mAP) of 94.6%, while physical grasping experiments attained a success rate of 93%. These findings confirm the effectiveness of the sensor–algorithm integration and demonstrate the robustness and practical feasibility of the proposed system.