Vision-Based Household Waste Classification and Smart Bin Development

The environment was negatively impacted because of the incorrect disposal of domestic waste. At the moment, the five most popular disposal methods are composting, waste compaction, landfilling, incineration, and biogas generation. In particular, landfills and incinerators are highly common in many countries; however, they may release certain pollutant gases contributing to the green house effect while there may be possibilities of waste generation. While classifying waste items is the key to waste management, household trash is usually classified incorrectly due to residents' negligence of the importance of waste classification, imperfect infrastructure, or the need for more instructions from local authorities. To address this problem, we propose a vision-based approach for waste classification and develop a prototype to test the system's real-time function. We applied transfer learning to fine-tune four well-known pre-trained object detection models on a self-collected dataset of 1100 samples. A smart trash bin (STB) prototype is developed for system validation and verification. The models are then implemented on the developed STB, and their CPU and GPU performance is tested. Experiment results have emphasized the effectiveness and feasibility of the proposed approach to address ecological problems in cities.