Implementation of Deep Learning for Organic and Anorganic Waste Classification on Android Mobile

Ramadhani, R.D and Thohari, A.N.A and Kartiko, C and Junaidi, A and Laksana, T.G (2021) Implementation of Deep Learning for Organic and Anorganic Waste Classification on Android Mobile. Proceedings of The International Conference on Innovation in Science and Technology (ICIST 2020). ISSN 2352-5401

	Text (Dokumen Lengkap (Sampul, Informasi Dewan Redaksi, Editor/Panitia Pelaksana, Daftar Isi, Artikel))) Lengkap - Implementation of Deep Learning for Organic and Anorganic Waste Classification on Android Mobile.pdf Download (1MB)
	Text (Bukti Korespondensi (Submission, LOA, Invitation)) Koresponden ICIST.pdf Download (318kB)
	Text (Sertifikat ICIST 2020) III.A.2.a.2 Seminar International pada IOP Proceedings_4-4.pdf Download (115kB)

Abstract

In this paper, a deep learning algorithm based on convolutional neural network (CNN) is implemented using pyhon and tensorzow lite for image classiycation on mobile. A large number different images which contains two types of waste, namely organic and anorganic are used for classiycation. The yrst stage to make classiycation model is prepare a dataset such as organic and anorganic waste images. Next divide both image in the training and validation directories. The split percentage when divide image is 90 percent for training and 10 percent for validation. After get image for training and testing, the next step is image augmentation to create new data from existing data. Next pre-processing using image data generator prepare the training data that will be implemented by the model. The important step in this process is make architecture of the CNN. In this paper used four layer convolution and there are two attributes that added to increase the accuracy of the training model. The yrst attribute is the dropout which make model become good yt and reduces overytting. The second is adding padding and stride attributes to speed up the step of epoch during training. So that, by using padding and stride make the training time 50 percent faster than before. After got model with accuracy more than 90 percent, the last step is testing model using image in validation directories. Based on testing step, model has been able to classify images of organic and anorganic waste correctly. Application can running smoothly and could classify waste using live camera or photo in gallery

Actions (login required)

View Item