A new machine vision-based industrial robot system for garbage management

Abstract

In response to challenges such as low efficiency, slow processing speed, and potential harm to human operators, this study presents the design and implementation of a machine vision-based garbage sorting system. The system integrates a robotic arm with advanced machine vision technologies to enhance sorting performance. Initially, a garbage sorting system is constructed around a robotic arm framework. Subsequently, the study evaluates various network architecture models and develops a neural network using MobileNetV3 and YOLOv4 (You Only Look Once version 4) by optimizing the backbone network components. The optimal grasp of the mechanical claws is determined using a method and minimum external torque algorithm, enabling the robotic arm to autonomously execute garbage sorting and classification tasks. Experimental results demonstrate that the system achieves a sorting accuracy of 90% for individual garbage items with the target category neural network and an overall classification accuracy of 91.2%. In scenarios involving multiple, non-adhering waste items, the system consistently maintains a 100% sorting rate. Even when dealing with adhesive or stacked waste items, the neural network’s target detection capabilities remain operational, although the accuracy of sorting success may decrease. This research validates the feasibility and reliability of the machine vision-based garbage sorting system through comprehensive experimental evaluations.