Figure: Mask R-CNN implemented on a piled of trash in the street


Improper disposal of solid wastes could lead to disruptive effects that impact daily human lives. Waste management serves a great importance especially to the health of human lives. To contribute to the effort in solid waste segregation, we created a machine learning model that could classify the trash images into two categories, namely: biodegradable and non-biodegradable. Trash images is inputted to the pre-trained Mask R-CNN where the output segmented images are categorized in a classification machine learning model. For the trash classification machine learning model, Xception model achieved the highest validation accuracy of 92.61%.

The resulting classification is fed back to the pre-trained Mask R-CNN for it to output the bounding box of the object and its classification. This model was implemented on both input image and video which will serve a great help to industrial waste management facilities. However, the trash should be properly spread out to reveal more information to the object detector and further improve the model’s chances of classifying the garbage correctly.