Gunardi, Syafwal (2026) Deteksi Gambar Buatan Ai Menggunakan Transfer Learning Mobilenetv2 dan Tensorflow Lite Pada Aplikasi Android. Other thesis, Universitas Islam Riau.

Text skripsi_213510057_watermark.pdf - Published Version Restricted to Registered users only Download (3MB) | Request a copy

Abstract

The rapid advancement of artificial intelligence technology, particularly Generative Adversarial Networks (GAN), has enabled the creation of highly realistic synthetic images that are difficult to distinguish from real ones. This phenomenon raises concerns regarding the spread of disinformation and visual hoaxes. Meanwhile, conventional Deep Learning detection models generally require high computational resources, making them inaccessible for mobile devices. This study aims to develop an efficient and accurate AI-generated image detection system on the Android platform using the MobileNetV2 Convolutional Neural Network (CNN) architecture with Transfer Learning techniques. The model was trained using a combined dataset of 4,000 images consisting of real and StyleGANgenerated faces, as well as digital art. For performance optimization on mobile devices (on-device), the model was converted using the TensorFlow Lite framework. Test results show that the model achieved an accuracy of 88% on the test data. Implementation on an Android device (POCO F6) demonstrated high efficiency with a final model size of 9.07 MB and an average inference speed of 17 milliseconds per image. Furthermore, the User Acceptance Testing (UAT) obtained an average score of 4.15, indicating that the application is well-received by users as a practical tool for verifying the authenticity of digital images.

Actions (login required)

View Item