Deep Learning Based Multi-Class Waste Object Detection and Classification

The growing waste problem caused by modern consumption patterns has created serious environmental and public health challenges. In response, this study has introduced an automated system designed to detect and classify municipal solid waste using advanced deep learning algorithms. A custom dataset of 726 images, each image containing multi-class waste objects namely plastic, metal, organic, fabric, e-waste, paper and cardboard was created for training and evaluation purposes. The methodologies include testing deep learning models like Attention-based Long Short-Term Memory (AttLSTM), Convolutional Neural Networks (CNN), CNN combined with Vision Transformers (CNN+ViT), Faster Region-based CNN (Faster RCNN), Multiple Paths CNN (MPCNN), Single Shot Multibox Detector (SSD), and You Only Look Once version 9 (YOLOv9). Mean average precision (mAP) was used to assess the performance of the models. YOLOv9 came out on top with an impressive mAP of 97.23%, while Faster R-CNN followed with a mAP of 85.31%. The CNN+ViT and SSD models achieved mAP scores of 56.96% and 76.16%, respectively. The findings show how effectively YOLOv9 and Faster R-CNN address the complex issue of multi-class waste classification. By adopting deep learning techniques, identification and classification of waste management can be accelerated and efficiency can be achieved. Over time, this approach could lead to better waste disposal practices, benefiting both the environment and public health.