Optimizing household waste sorting through iterative learning using: YOLOv5 and YOLOv7

Abstract

Efficient home trash management is crucial for sustainable urban living, particularlyin densely populated areas like Dhaka City, Bangladesh. This thesis exploresoptimizing garbage sorting processes using iterative learning by comparingtwoadvanced object detection models, YOLOv5 and YOLOv7. The primary objectiveisto develop a robust object identification system to recognize and categorizeabandoned waste into seven categories: plastic, biological, cardboard, clothing, glass, metal, and paper. To achieve this, secondary data from Kaggle and primary data fromthe Dhaka waste plant were utilized. The models were trained and evaluated on thesedatasets, with performance metrics indicating YOLOv7 achieved a higher accuracyof 97.5% compared to YOLOv5's 96.2%. These results demonstrate YOLOv7's superior capability in accurately detecting and classifying waste materials, making it promising tool for enhancing waste management systems. This research contributes to waste management by offering a comparative analysis of contemporary deep learning models and insights into their practical applications in urban environments. The findings underscore YOLOv7's potential to significantly improve the efficiency and accuracy of household waste sorting, thereby supporting more effective waste management strategies in Dhaka City.