A Mobile Based Waste Classification Using MobileNets-V1 Architecture

Abstract

In Indonesia, waste is a very serious problem. According to research in handling and processing waste is classified into three types, namely recyclable, nonorganic, and organic waste. Organic and inorganic waste will generally be transported and stockpiled at the Final Disposal Site (TPA). So far, the only available waste bins are waste bins with manual sorting done by the community. As it is known that currently there are many people who do not understand the different types of waste to be disposed of, so that even though organic and inorganic types of waste have been provided, people still dispose of waste in inappropriate types. This of course will be very inconvenient in the effort to sort waste in the waste whereas the first place for garbage to gather. Because of the need for a tool that can help the community in distinguishing the types of waste before putting it into the waste with an accurate classification method Based on the problems in classifying the types of waste that have been described previously, we need a system that is able to classify waste according to its type, The MobileNets-V1 architecture is used in this research to classify images. The models generated by the architecture will then be deployed into mobile-based applications. The dataset used in this study consists of 3 classes, namely N (Non-Recyclable), O (Organic), R (Recyclable). Because the data is highly imbalanced, we conduct undersampling in order to balance the data. This undersampling process is done only in the training set after splitting the whole dataset into training, validation, and testing set. After the balancing process, each class has 1822 sample data, totalling of 5466 sample data in the trianing set. The pretrained MobileNets-V1 model is able to classify types of waste very well. The best model obtained is a model that uses dropout value of 0.4 which provides testing accuracy of 88.26%, training accuracy of 92.44% and validation accuracy of 89.00%.