Study of Visual Classification Transport Machines Based on Residual Networks and YOLOv8
Yingkun Hu, Xiufang Zeng, Mingcong Liu, Chenmei Ai, Congling Hou · 2025
We proposes a waste recognition and grabbing transportation robot composed of key components such as a dual-camera system, a three-axis robotic arm, and an Automated Guided Vehicle (AGV). Based on the captured images, the robot first performs waste image extraction through threshold segmentation in the HSV color space and contour area thresholding. It then designs a waste classification residual neural network using transfer learning and the addition of an attention mechanism. Finally, the AGV performs visual path tracking combined with YOLOv8s digital landmark recognition to complete the waste transportation task. This system saves time and effort, avoids direct human contact with waste, improves the accuracy of waste classification, and significantly reduces the cost and convenience of waste sorting. Experimental results show that the model has good robustness and is lightweight enough to be directly loaded onto a Raspberry Pi. The recognition accuracy for waste types exceeds 99%, and the robot can accurately locate the waste, with a waste grabbing success rate above 95%. The accuracy of digital landmark recognition reaches over 96%, indicating that the system has the capabilities of automatic waste recognition, classification, automatic grabbing, and transportation to designated areas.