WasteScan: Efficient Waste Detection and Multi-label Classification Leveraging Object Detection Methods

Abstract

This study presents an efficient waste detection and multilabel classification algorithm using single-shot object detection techniques. The objective is to accurately identify and classify various waste objects according to their material, even in scenarios in which objects are densely clustered. The proposed algorithm demonstrates promising outcomes, achieving both exceptional detection accuracy and efficient computational performance. To achieve efficient waste detection, we utilize the YOLOv5, YOLOv7 and YOLOv8 models, known for their ability to detect objects with high precision. These models utilise a one-shot detection technique to predict bounding boxes and class probabilities for many objects in a single run, enabling real-time image processing. This study explored the effectiveness of an incremental learning approach, showcasing notable performance gains for a newly added class. This characteristic is especially crucial for applications that require timely decision-making or monitoring waste in dynamic environments. Furthermore, the models' computational speed guarantees real-time performance, which qualifies them for waste management applications which involve real-time sorting functionalities.