Developing a plogging activity tracking app using deep learning for image recognition

Abstract

Plogging is an activity that combines jogging with picking up litter, and participants often share their efforts on social media. However, the repetitive bending involved in plogging may cause back strain, and manually entering details such as the location and quantity of litter could slow the spread of this activity. This study sought to create and test a deep learning application to automatically monitor and record plogging by identifying the type and quantity of litter. We employed Convolutional Neural Networks (CNN) and YOLOv5 to develop an image recognition model. This model allowed users to easily log their plogging efforts by simply taking a photograph, removing the need to manually input the litter details. Moreover, we proposed a reward system that uses the collected trash amount and the distance covered to promote competition among users. We developed the first application that uses deep learning to automatically identify litter for tracking plogging activities. However, as this application was only a prototype, no comparative studies or usability tests were done. In future research, we plan to assess the application's usability and compare it with other similar applications.