Advanced Waste Detection Leveraging YOLO for High-Precision Classification

This study aims to investigate the impact of improper waste disposal, addressing a critical gap in the existing literature and providing a sustainable waste management solution incorporating Artificial Intelligence. The research seeks new insights into advanced garbage detection systems. An open source pre-existing annotated dataset consisting of 3501 training set of images, 348 validation set of images and 228 test set images of a single class from roboflow was used to train all the models on 12 different versions of YOLO (You only look once) and 2 different versions of Gelan Series. All models were trained on an NVIDIA A10g 24GB GPU for 50 epochs, applying data augmentation techniques like rotation, cropping, and blurring. Among all models, YOLOV9-c and YOLOV8m achieved the best performance. Additionally, a full-stack application was developed as a Proof Of Concept using Flutter for the front end and Django for the back end.