Industrial Artificial Intelligence Methods for Fabric Color-Defect Detection in Textile Manufacturing

Abstract

Color-related fabric defects such as shade variation, off-shade, discoloration, and color bleeding remain challenging to detect consistently in textile manufacturing because camera-based inspection is highly influenced by illumination variability, fabric reflectance, and domain shift across production batches. These limitations often reduce the reliability of automated inspection systems and highlight the need for a systematic synthesis of existing Industrial Artificial Intelligence approaches. Objective: This study aims to systematically review and synthesize Industrial Artificial Intelligence methods for fabric color-defect detection to identify methodological trends, evaluation practices, and research gaps, as well as to explain why these approaches are important for improving reliability in industrial quality control. Method: The research employs a Systematic Literature Review design, collecting peer-reviewed studies through structured database searches followed by PRISMA-guided screening and eligibility assessment. The selected studies are analyzed using comparative narrative synthesis and standardized coding of method types, color representation, illumination handling strategies, datasets, and evaluation metrics. Findings: The review reveals four dominant methodological groups: classical computer vision, supervised deep learning, reconstruction-based anomaly detection, and feature-space anomaly or hybrid approaches. Across these approaches, robust performance consistently depends on a triadic design principle consisting of color-consistent representation, illumination robustness, and learning strategies aligned with label availability. The study also identifies a key evaluation gap where conventional vision metrics are rarely complemented by perceptual color-difference measures. Implications: The findings suggest that future research and industrial implementation should focus on developing color-calibrated datasets, adopting dual-axis evaluation frameworks that include perceptual color metrics, and validating models under varying illumination and fabric conditions to enhance real-world reliability. Originality/Value: This study provides an original contribution by proposing a color-defect–focused operational framework that integrates color science principles with Industrial AI method selection and deployment constraints, offering clearer guidance than previous reviews that primarily addressed structural textile defects.