Automated Fabric Defect Detection Techniques

Automated fabric defect detection has evolved from early rule-based and statistical methods to advanced machine learning approaches that ensure consistency, speed and precision in textile quality control. Initial techniques relied on template matching, edge detection and texture analysis to identify irregularities in weave patterns. The advent of convolutional neural networks enabled end-to-end learning of defect features, while unsupervised models have reduced dependence on labelled faulty samples. Recent innovations incorporate generative augmentation, attention-guided feature fusion and multi-scale analysis to improve sensitivity to subtle flaws across diverse fabric types. These systems deliver real-time inspection capabilities on production lines, minimise waste and support global sustainability goals by reducing resource loss and enhancing product reliability.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

One line of work introduces a mask-guided diffusion framework that augments scarce defect samples by learning the correlation between background texture and damage shape. An attention-guided network then fuses high-level semantic cues with low-level details to boost detection metrics while halving model complexity. Another approach employs a multi-scale convolutional denoising autoencoder to reconstruct defect-free patches and uses reconstruction residuals at several pyramid levels as anomaly indicators. This unsupervised scheme requires only normal samples for training, achieving robust localisation across simple and patterned fabrics. A third study enhances a popular one-stage detector by integrating a novel spatial pooling block and soft-pool operators, coupled with adaptive histogram equalisation for contrast normalisation. The improved model attains a substantial increase in mean average precision with minimal loss of throughput, demonstrating its suitability for real-time industrial deployment.

Automated Fabric Defect Detection Techniques publication trend

The graph below shows the total number of articles in automated fabric defect detection techniques across all publications each year (not limited to Nature Index journals).

Technical terms

Diffusion model: A generative framework that iteratively adds and removes noise to learn data distributions and produce realistic synthetic examples.

Denoising autoencoder: An unsupervised neural network that learns to reconstruct clean inputs from noisy versions, highlighting anomalies as reconstruction errors.

Attention mechanism: A module that weights feature importance across spatial or channel dimensions, enabling networks to focus on salient defect characteristics.

You Only Look Once (YOLO): A real-time, single-shot object detection algorithm that divides an image into grids to predict bounding boxes and class probabilities in one pass.

Multi-scale fusion: The integration of features extracted at different spatial resolutions to capture both global context and local detail for precise defect localisation.

References

1. MED-AGNeT: An attention-guided network of customized augmentation of samples based on conditional diffusion for textile defect detection. International Journal of Cognitive Computing in Engineering (2025).
2. Automatic Fabric Defect Detection with a Multi-Scale Convolutional Denoising Autoencoder Network Model. Sensors (2018).
3. Fabric defect detection using the improved YOLOv3 model. Journal of Engineered Fibers and Fabrics (2020).