Development of Computer Vision Based Aluminium Scrap Sorting Using Real Time Metal Scrap Dataset

The aluminium manufacturing industries face numerous challenges when it comes to sorting metal scraps, particularly when scraps are small in size. Traditional methods like flotation and gravity separation are inefficient, energy-intensive, and often fail when dealing with size variations. Recent advancements in computer vision and deep learning algorithms, including intelligent robotic systems, offer promising solutions. However, challenges such as accurate detection in complex environments, varying lighting conditions, and overlapping objects still persist. Newer techniques like YOLO (You Only Look Once) for real-time object detection, and Transfer Learning to improve model accuracy with smaller datasets, have shown potential. This research aims to overcome these challenges by developing an AI-powered, computer vision-based sorting system that can identify and classify aluminium scraps by color and shape at high speed. The authors developed a custom dataset and trained models using CNN architectures such as DenseNet-161, ResNet-152, and VGG 19. By employing ensemble techniques, this research achieved an 75% of accuracy.