Garbage Image Segmentation Using Combination of Thresholding Algorithms and Pyramid Scene Parsing Network

Abstract

Garbage is one of the biggest problems in Indonesia, even most of country in the world. The amount of garbage increases every year, therefore a reliable garbage management system is needed to prevent the contamination of garbage on environment. Garbage sorting (based on its material) is an important part of garbage management system. Since the amount of garbage is really high, a fast and accurate garbage sorting process is needed to improve the overall garbage management system. One of the most important tasks in garbage sorting process is garbage image segmentation. This research proposes garbage image segmentation process by using Pyramid Scene Parsing Network (PSPNet) and three binary images from the combination of thresholding algorithms as the input for the network. Then it is compared with PSPNet that uses a RGB image as an input. The results show that when using PSPNet, RGB image is better than the image from combination of thresholding algorithms as an input for PSPNet with the maximum F1 Score is up to 98%. However, the results of this comparison are competitive where the difference of F1 score between the proposed method and PSPNet with RGB image is less than 0.02.