Deteksi Defect Sepatu Berbasis Fine-Tuned CNN Menggunakan Arsitektur DenseNet untuk Mendukung Proses Quality Control

Abstract

Indonesia is a leading global exporter of footwear. Despite successfully exports million pair of shoes, shoes industry in Indonesia faces significant challenges due to conventional quality control (QC) limitations. This study proposes a deep learning-based defect detection model to enhance the efficiency and accuracy of the QC process. This study uses primary datasets collected from Laboratory of PIDI 4.0 Politeknik ATK Yogyakarta and employed in four experimental scenarios. The dataset consists of 2,828 images comprising of two classes (good class and reject class). This study utilizes fine-tuned CNN using DenseNet architecture for distinguishing two classes. Before employing the fine-tuned CNN method, we separate the data into three components, which are training, validation and testing data, with proportion of 80:10:10. The results demonstrate that the integration of data augmentation and the Adam optimizer yielded the highest performance, achieving accuracy, precision, recall and F1-score of 0.9546, 0.9551, 0.9546, 0.9545, respectively. These findings suggest that implementing automated deep learning models has potential to reduce rejection rates by modernizing traditional inspection methods, which can strengthen the competitiveness of Indonesia’s footwear industry in the global market.