Towards Cleaner Environments by Automated Garbage Detection in Images

Abstract

The environment protection is becoming, now more than ever, a serious consideration of all government, non-government, and industrial organizations. The problem of littering and garbage is severe, particularly in developing countries. The problem of littering is that it has a compounding effect, and unless the litter is reported and cleaned right away, it tends to compound and become an even more significant problem. To raise awareness of this problem and to allow a future automated solution, we propose developing a garbage detecting system for detection and segmentation of garbage in images. For this reason, we use deep semantic segmentation approach to train a garbage segmentation model. Due to the small dataset for the task, we use transfer learning of pre-trained model that is adjusted to this specific problem. For this particular experiment, we also develop our own dataset to build segmentation models. In general, the deep semantic segmentation approaches combined with transfer learning, give promising results. They show great potential towards developing a garbage detection application that can be used by the public services and by concerned citizens to report garbage pollution problems in their communities.