Frequency-Domain Analysis of Textile Fabric Defects Using the Central Spatial Frequency Spectrum Method

Automated defect detection in woven textile fabrics remains a persistent challenge in modern spinning and weaving mills. Human visual inspection—the prevailing industrial standard—achieves detection rates of only 60–75%, owing to the repetitive, low-contrast, and sub-millimetre nature of common fabric defects. Conventional intensity-based machine vision systems, while superior to human inspection, fail to capture the structural signature of defects that manifest as disturbances in the periodic architecture of woven fabric. This paper presents a rigorous analysis of a paradigm-shifting approach: the Central Spatial Frequency Spectrum (CSFS) method, which re-formulates fabric defect detection as a frequency-domain signal processing problem. By applying the two-dimensional Discrete Fourier Transform (2D-DFT) and reducing the spectral representation to two orthogonal central slices—F(fx, 0) and F(0, fy)—along the warp and weft directions, the method extracts seven discriminative physical parameters (P₁–P₇) that serve as a compact feature vector for defect classification. The Fast Fourier Transform (FFT) algorithm reduces the computational complexity from O(N²) to O(N log N), enabling real-time online inspection at loom speeds. The CSFS method achieves a detection accuracy of over 91%, significantly outperforming intensity-based methods (~74%) and human inspection (60–75%). Applications to double-yarn, missing-yarn, broken-fabric, and density-variation defects are demonstrated with their unique frequency fingerprints. Implications for spinner productivity, quality loss reduction, and integration with modern AI-based inspection platforms are discussed. Practical limitations—particularly applicability to periodic versus non-periodic fabric structures—are critically evaluated.