Secretary bird optimised support vector machine model for adire fabric defect detection

Abstract

Adire fabric is a traditional hand-dyed fabric indigenous to the Yoruba people of Nigeria. It is produced using manual resist-dyeing techniques involving materials such as raffia, wax, starch, or stitching. These handcrafted processes while culturally significant, make the fabric highly susceptible to a variety of defects such as stains, tears, pattern misalignment, colour variation, colour crocking, sulphur-mark and, colour-smear. Traditional visual inspection methods are labour-intensive, time consuming and prone to error. Hence, this study developed an optimised Support Vector Machine (SVM) model using the Secretary Bird Optimisation Algorithm (SBOA) for automated detection of Adire fabric defects. 234 Adire fabric images were acquired using a Redmi 14C 50MP digital camera and augmented to 884. Among the data collected, 396 images represented various type of defects while 488 were non-defective (Normal). Preprocessing involved Gaussian filtering and Contrast Limited Adaptive Histogram Equalisation (CLAHE), while Gray-Level Co-occurrence Matrix (GLCM) was used for texture-based feature extraction. SBOA was applied to optimise SVM hyperparameters (penalty factor , kernel type, gamma (  ), and polynomial degree) yielding the SBOA-SVM model, implemented in MATLAB R2023a. The performance of the model was evaluated against standard SVM using accuracy, sensitivity, specificity, false positive rate as metrics. SBOA-SVM outperformed standard SVM across all defect categories. For colour crocking and colour smear, SBOA-SVM achieved accuracy of 95.81% and 96.04%, and sensitivity of 85.29% and 85.61%, respectively, while the corresponding values obtained for standard SVM were 93.55% and 94.23%. Challenging defects like colour variation, tear, and stain, SBOA-SVM improved sensitivity to 70.45%, 72.73%, and 72.50% at specificity above 98%. The developed SBOA-SVM model demonstrated superior accuracy and efficiency, establishing its potential for automated defect detection in Adire and other related textile applications.