Deep Augmented Object Detection for Waste Material Recognition

Waste material identification is important for the purpose of environmental sustainability, maximizing the effectiveness of waste management, and facilitating automatic cleanup mechanisms. In this research, the application of YOLO models—namely, YOLOv5 to YOLOv12, and YOLO-World—is explored with the aim of detecting waste materials in different real-world images. A custom waste detection dataset is used, divided into training, validation, and testing sets, for carrying out extensive performance evaluation. The paper contrasts pre- and post-augmentation performance using data augmentation techniques devised to improve model accuracy and generalization. The performance is quantified by precision, recall, mAP@50, and mAP@50–95. Before augmentation YOLOv9 was 88.8% the top-performing model and YOLOv6 was 73.1% the lowest-performing model. After augmentation implementation, YOLOv9 achieved the best detection accuracy of 89.3%, and YOLOv6 achieved the lowest accuracy of 76.2%. These findings validate that augmentation greatly improves detection performance and demonstrates the potential of computer vision applications in waste management contexts.