Penerapan Metode CNN (Convolutional Neural Network) untuk Mengklasifikasikan Jenis Cacat pada Kulit Hewan

Eka Legya Frannita (Politeknik ATK Yogyakarta, Yogyakarta)
Alifia Revan Prananda (Universitas Tidar, Magelang)

TIN: TERAPAN INFORMATIKA NUSANTARA
Vol 5 No 2 (2024): July 2024

Recently, leather industry was rapidly growth in several countries. In Indonesia, leather industry became one of the government's priority industries since there were quite a lot of leather industries developing in various regions in Indonesia. On the other hand, there were large number of consumer demand for leather products. Regarding to this fact, maintaining the quality of leather was strongly important. An alternative solution for maintaining leather quality is to conduct leather quality inspection process. However, currently the leather inspection process was still carried out manually by identifying directly the types of defects found on the surface of the leather. This manual inspection process certainly has several hurdles such as time consuming, requiring high accuracy, and requiring experienced operators. This research aimed to develop convolutional neural network architecture that can classify types of leather defects. This research was done by conducting four main processes which were literature study and data collection processes, develop CNN architecture, training process, and testing process. This research work used public dataset consisting of 3600 digital leather images distributed into six classes (folding mask, grain off, growth marks, loose grains, pinhole, non-defective). Based on the training and testing process, the model obtained training accuracy of 90.43% and testing accuracy of 88.47%.