CD Skripsi

Implementasi Algoritma Convolutional Neural Network Untuk Identifikasi Alfabet Sistem Isyarat Bahasa Indonesia Menggunakan Arsitektur Mobilenet Dan Metode Ensemble

---

This research focuses to identify alphabets in the Sistem Isyarat Bahasa Indonesia (SIBI) using the Convolutional Neural Network (CNN) algorithm with a transfer learning approach. Three CNN architectures were used: MobileNetV1, MobileNetV2, and MobileNetV3-Small. The dataset consists of 1,560 static hand gesture images representing 26 alphabet letters, collected from public sources (Kaggle) and manual documentation. All images were processed through augmentation and resized to 224×224 pixels before training. The models were trained using pre-trained weights from ImageNet and further optimized through fine-tuning. The results showed that MobileNetV1 and MobileNetV2 performed consistently well, achieving validation accuracies of 91% and 88% respectively after fine-tuning. In contrast, MobileNetV3-Small only reached 40% accuracy and exhibited signs of severe overfitting, and was therefore excluded from the ensemble stage. The two best-performing models, MobileNetV1 and MobileNetV2, were combined using the ensemble majority voting method to improve prediction accuracy. The final results showed that the combined model achieved an accuracy of 92%, outperforming each individual model. These findings demonstrate that the ensemble method is effective in enhancing the performance of CNN-based SIBI alphabet identification.

Keywords: CNN, Ensemble Voting, Image Classification, MobileNet, SIBI.

---

Ketersediaan

Informasi Detail

Judul Seri

-

No. Panggil

2103110234

Penerbit

Pekanbaru - : Universitas Riau FMIPA Sistem Informasi., 2025

Deskripsi Fisik

-

Bahasa

Indonesia

ISBN/ISSN

-

Klasifikasi

2103110234

Tipe Isi

-

Tipe Media

-

Tipe Pembawa

-

Edisi

-

Subjek

SISTEM INFORMASI

Info Detail Spesifik

-

Pernyataan Tanggungjawab

Mutia

Versi lain/terkait

Lampiran Berkas

Komentar

---

Anda harus masuk sebelum memberikan komentar