Tongaraas: Tongs for Recognizing Littering Garbage with Active Acoustic Sensing

Littering has developed into a serious environmental problem. However, the actual situation of litter and the results of litter pickup activities are not organized as information. Therefore, the objective of this research is to grasp the distribution of the type and location of litter comprehensively. To achieve the objective of this research, we have proposed a method for recognizing litter using an acoustic sensor on a smartwatch worn on the wrist and a method for recognizing litter using a small camera mounted on tongs. However, in these studies, there were limitations in the range of litter type estimation, lack of recognition accuracy, and privacy issues. To solve the above problem, we propose a litter type recognition system, named Tongaraas, that combines active acoustic sensing with tongs. In the evaluation experiment, we built the litter type recognition model for six categories of litter. The evaluation results showed the models, which were trained with dataset collected by single person and three people, perform at F-value of 0.978 (SVM) and 0.849 (LightGBM), respectively. It suggests it is possible to estimate with common litter type recognition model, although there is a certain level of negative effects due to the individual difference.