SKRIPSI IF
KLASIFIKASI CITRA MOTIF BATIK NUSANTARA MENGGUNAKAN ALGORITMA CONVOLUTIONAL NEURAL NETWORK (CNN) DENGAN MODEL RESNET50, MOBILENETV2, DAN INCEPTIONV3
ABSTRAK
Batik nusantara merupakan warisan budaya Indonesia dengan keragaman visual yang sangat tinggi. Namun, kemiripan pola, bentuk, dan warna antar motif batik sering kali menimbulkan kesulitan dan kesalahan dalam proses identifikasi secara manual. Penelitian ini bertujuan untuk membangun sistem klasifikasi citra motif batik nusantara berbasis deep learning serta membandingkan performa tiga arsitektur Convolutional Neural Network (CNN), yaitu ResNet50, MobileNetV2, dan InceptionV3. Dataset yang digunakan diperoleh dari Kaggle, mencakup 20 kelas motif batik dengan total 3000 citra dengan 150 citra per kelas. Dataset dibagi menjadi 3 dataset berbeda, yaitu data train sebanyak 2100 citra, validation sebanyak 440 citra, dan data test sebanyak 460 citra. Metode yang diimplementasikan adalah teknik transfer learning berbasis bobot awal ImageNet, didukung oleh on-the-fly data augmentation, penggunaan dropout, serta integrasi tiga callbacks yaitu EarlyStopping, ReduceLROnPlateau, dan ModelCheckpoint. Hasil komparasi menunjukkan arsitektur ResNet50 memegang performa terbaik dengan capaian nilai accuracy tertinggi sebesar 97% dengan waktu pelatihan 1 jam 15 menit 21 detik selama 14 epoch sebelum dihentikan oleh EarlyStopping. Sementara itu, MobileNetV2 menghasilkan accuracy sebesar 95,43% dengan durasi komputasi paling efisien yaitu 32 menit 40 detik, dan InceptionV3 menghasilkan accuracy 95% dengan total waktu pelatihan selama 2 jam 37 menit 42 detik. Secara keseluruhan, model ResNet50 dapat disimpulkan sebagai arsitektur paling optimal dibandingkan MobileNetV2 dan InceptionV3 pada penelitian ini.
Kata Kunci: Batik, Convolutional Neural Network, Image Processing, ResNet50, MobileNetV2, InceptionV3
ABSTRACT
Batik Nusantara is an Indonesian cultural heritage with highly diverse visual styles. However, the similarities in patterns, shapes, and colors across different batik motifs often lead to difficulties and errors in manual identification. This study aims to develop a deep learning-based image classification system for Indonesian batik motifs and compare the performance of three Convolutional Neural Network (CNN) architectures: ResNet50, MobileNetV2, and InceptionV3. The dataset used was obtained from Kaggle, covering 20 independent batik motif classes with a total of 3,000 images (150 images per class). The dataset was split into three different datasets: training data with 2100 images, validation data with 440 images, and testing data with 460 images. The implemented method is the transfer learning technique based on ImageNet pre-trained weights, supported by on-the-fly data augmentation, dropout layers, and the integration of three callbacks: EarlyStopping, ReduceLROnPlateau, and ModelCheckpoint. The comparative results show that the ResNet50 architecture achieved the best performance with the highest overall accuracy of 97%, requiring a training time of 1 hour 15 minutes 21 seconds across 14 epochs before being halted by EarlyStopping. Meanwhile, MobileNetV2 yielded an accuracy of 95.43% with the most efficient computational duration of 32 minutes 40 seconds, and InceptionV3 achieved an accuracy of 95% with a total training time of 2 hours 37 minutes 42 seconds. Overall, the ResNet50 model can be concluded as the most optimal architecture compared to MobileNetV2 and InceptionV3 in this study.
Kata Kunci: Batik, Convolutional Neural Network, Image Processing, ResNet50, MobileNetV2, InceptionV3
| S260908019 | 607.2 ANS k | Perpustakaan STMIK AMIKBANDUNG | Tersedia |
Tidak tersedia versi lain