Advanced Waste Detection: Mobilenetv2 SSD and Yolov8 for Waste Bank Mobile Application

The research addressed the challenge of waste management in waste banks using deep learning-based object detection. The study focused on comparing two models, MobileNetV2 SSD and YOLOv8, for identifying and classifying inorganic waste. The research process involved preparing a dataset of waste images, which were then used to train and evaluate models. The research methodology encompassed a systematic approach, commencing with preparation of a dataset comprising waste images. These images were then labeling and subsequently partitioned into distinct training, validation, and test sets, ensuring a robust foundation for model training and evaluation. The training and evaluation phase leveraged key performance metrics, namely mAP (mean Average Precision) and AR (Average Recall), to gauge the efficacy of both models. The results unequivocally demonstrated the consistent superiority of the YOLOv8 model over MobileNetV2 SSD across these evaluation metrics. Notably, YOLOv8 achieved an impressive mAP of 0.834 and an $A R$ of 0.791, underscoring its exceptional accuracy and stability in waste detection. The research conclusion is that YOLOv8 stands as the preferred model for seamless integration into waste bank mobile applications, owing to its enhanced detection capabilities. These advancements significantly bolster recycling efforts and promote heightened awareness and participation in waste management practices.