Development of an automatic garment defect recognition system using a machine learning approach

Abstract

Quality assurance in garment manufacturing remains a critical challenge for developing-country exporters, particularly in Ethiopia's emerging apparel sector where manual visual inspection predominates. This subjective, fatigue-prone process leads to inconsistent defect detection and limits production throughput. This study presents an automated garment defect recognition system integrating a VGGNet-19 convolutional neural network (CNN) for feature extraction with an artificial neural network (ANN) for multiclass classification. A dataset of 1000 trouser images encompassing four defect categories (broken stitching, open seams, seam slippage, and untrimmed threads) and defect-free samples was collected from Addis Ababa garment manufacturing facilities under factory lighting conditions. The proposed architecture achieved an overall classification accuracy of 86.7% on the held-out test set, with per-class performance of 86.7% for defect-free, 83.3% for broken stitching, 90.0% for open seams, 80.0% for seam slippage, and 96.7% for untrimmed threads. The system processes images in 0.45 seconds per sample on consumer-grade hardware, representing a substantial 60-fold improvement over manual inspection throughput while maintaining consistent defect detection criteria. Comparative analysis revealed that RGB color channels outperform grayscale conversion (86.7% vs. 80.0% accuracy), and the deep VGGNet-19 architecture exceeds the shallower AlexNet by 6.2 percentage points despite longer training requirements (6.2 hours vs. 3.5 hours). This work establishes a foundation for real-time, automated quality control in resource-constrained apparel manufacturing environments, with a return on investment estimated at 12–18 months for medium-sized Ethiopian factories.