Garbage classification using Depthwise Separable Convolution with data augmentation
Budi Dwi Satoto(a*), Achmad Yasid(a), Faroid(a), Aghus Setio Bakti(a), Muhammad Yusuf(a), Budi Irmawati(b)

(a) Information system department
University of Trunojoyo Madura
Bangkalan, East Java, Indonesia
budids[at]trunojoyo.ac.id
(b) Informatics Engineering Department
University of Mataram
West Nusa Tenggara, Indonesia
budi-i[at]unram.ac.id

Abstract

Tourism destinations are often beautiful and valuable natural areas. Good waste management helps maintain the cleanliness and beauty of the natural environment, minimizes negative impacts on the ecosystem, and ensures the sustainability of tourist destinations. Additionally, waste management creates opportunities for the recycling industry. By separating, collecting, and processing recyclable waste, such as paper, plastic, metal, and glass, this industry creates jobs and produces products that can be sold. Tools are needed to facilitate visually sorting waste in tourism areas. It can be done with the help of deep learning. The contribution is to use a combination of Depthwise Separable Convolution architectural concepts, hoping that computing will be lighter, maintain accuracy, and remain stable. The model is relatively small, which makes it suitable for mobile devices with limited computing power and storage. The dataset consists of six classes: Cardboard, glass, metal, paper, plastic, and trash. Because of data limitations, augmentation techniques are used. The test results show an average model accuracy of 98.29% with a training computing time to obtain a model of 45 minutes. MSE 0.0343, RMSE 0.1852, and MAE 0.0229. Testing with new experimental data takes an average of 1-2 seconds

Keywords: Garbage classification, Deep learning, Depthwise Separable Convolution, Data augmentation

Topic: Artificial Intelligence and Data Science