Convolutional Neural Network with VGG-16 Architecture for Object Image Classification in Arabic Language

• Published in: International Conference on Wireless and Telematics (Online) pp. 1 - 6
• Main Authors: Ramadhan, Iqbal Putra, Maylawati, Dian Sa'Adillah, Ramdania, Diena Rauda, Wahana, Agung, Slamet, Cepy, Fuadi, Rifqi Syamsul
• Format: Conference Proceeding
• Language: English
• Published: IEEE 04.07.2024
• Subjects: Accuracy; Convolutional Neural Network; Data models; Deep Learning; Object Detection; Programming; Telematics; Training; Training data; Wireless communication
• ISSN: 2769-8289
• DOI: 10.1109/ICWT62080.2024.10674728

Arabic is one of The challenging languages to learn. It takes 1320 intensive hours to reach a proficient level. The media plays an active role in enhancing learning and achieving students' goals, which significantly influences the success and efficiency of education, especially for school students who find it challenging to learn the Arabic language. Based on this, a classification model is needed to interpret objects in Arabic, making The learning process more efficient and varied. The design of this system uses The Convolutional Neural Network (CNN) method as an image data processor. All research data uses image pictures as input and output, including Tables, Pens, Chairs, Whiteboards, Clocks, People, Mobile Phones, Plants, Books, and Bags. All research data consists of 7,341 images of class objects, including training and testing data. The training data uses 6085 images, and the test data uses 1256 images each. This research employs four testing scenarios with variations in learning rate and dropout parameters. Learning rate variations include 0.001 and 0.0001, while dropout variations are 0.3 and 0.5. The test results from all scenarios show that a learning rate of 0.001 and a dropout of 0.5 have The best accuracy value at 90%. This value proves that The CNN algorithm performs well in detecting class objects.