Design of an Intelligent Waste-Sorting Robot Based on Image Recognition

Abstract

The growing waste crisis in the world at large requires automated sorting to improve on recycling efficiency and advance the application of the circular economy. In the paper, the developed intelligent waste-sorting robot combines the deep learning-based on YOLOv8 with robot manipulation systems to classify waste and sort it automatically. The given system is implemented with the usage of an RGB-D camera to obtain images, noise reduction and normalization preprocessing algorithms, and a YOLOv8 neural network to detect multiple classes of objects in 9 categories of waste. A 6-DOF collaborative robotic arm with adaptive gripper performs physical sorting tasks computed on the basis of the 3D coordinates and motion planning algorithms. Using an experimental dataset of 8,500 annotated images, the 94.3% mAP at 0.5 detect rate, 92.5% precision, 91.1% recall, and 8.7 ms inference time are achieved, which makes it possible to process images at 115 frames per second in real-time. Robotic manipulation module has 92.3 percent grasping success and the system throughput is 445 objects/hour that is impressive when compared to manual sorting techniques. The findings confirm the feasibility of the system in practice in two areas: municipal recycling centers and industrial waste management and also in niche applications such as e-waste processing.