JURNAL TEKNIK INFORMATIKA DAN SISTEM INFORMASI
Vol 8 No 2 (2021): JATISI (Jurnal Teknik Informatika dan Sistem Informasi)

Klasifikasi Genre Manga Menggunakan Convolutional Neural Network

Reynaldo Stefanus Hartono (Unknown)
Hendrawan Armanto (Unknown)



Article Info

Publish Date
17 Jun 2021

Abstract

Researcher conducted experiments about manga genre classification based on input in the form of black and white image and manga cover. Assigning genres to manga is a particularly challenging task because genre is a feature that is not physically present in a movie frame, so off-the-shelf image detection models cannot be easily adapted to this context. Moreover, multi-label classification is more challenging than binary classification or multi-class classification considering that one instance can be assigned to multiple class at once. After successfully making genre predictions, then what needs to be done is to analyze the existing data so that it finds the factors that can affect the final results found. Convolutional Neural Network is a method that is often used to classify images because it is able to detect the main features of an image, the combination of architectural layers is able to produce complex and relevant predictions. The neural network architecture layer includes the convolutional layer, the pooling layer, and the fully connected layer. Each layer has parameters that can be changed according to research needs. Changes in these parameters allow researcher to obtain different results. The experiment was carried out in two stages. The first stage contains 15 experiments with the same convolutional neural network model structure. The second stage contains 24 experiments by changing the structure of the convolutional neural network model. The total number of experiments conducted was 39 trials. The best result obtained was the 38th trial with an accuracy of 46.3 percent and the number of active F1s was 25 genres. In addition, researchers have analyzed several factors that affect the results of the research that have been obtained in the hope that it will be useful for other researchers who want to carry out similar research.

Copyrights © 2021






Journal Info

Abbrev

jatisi

Publisher

Subject

Computer Science & IT

Description

JATISI bekerja sama dengan IndoCEISS dalam pengelolaannya. IndoCEISS merupakan wadah bagi para ilmuwan, praktisi, pendidik, dan penggemar dalam bidang komputer, elektronika, dan instrumentasi yang menaruh minat untuk memajukan bidang tersebut di Indonesia. JATISI diterbitkan 2 kali dalam setahun ...