Adaptive neuro-fuzzy system for quantitative evaluation of woven fabrics’ pilling resistance

Fabric pilling is considered a performance and aesthetic property of the woven products that determine its quality. The subjective evaluation of the fabric pilling results in misleading values that depend on the measurement standard even for the same sample. This work utilizes some textural features extracted from the fabric’s images to obtain better representative and quantitative values of the fabric’s surface. An algorithm for creating features dataset for training and testing the soft-computing classifier was described where random noise was added to the limited number of fabric’s pilling standard images. The objective pilling classification of the fabric samples was performed using an adaptive neuro-fuzzy system (ANFIS) which showed an ability to classify the noised standard images with a correct classification rate of 85.8%. The ANFIS was also able to classify actual fabric samples with a Spearman’s coefficient of rank correlation at +0.985 when compared with the classification grades of the human operators. Results showed high efficiency of the system that is independent on the different fabric structure or color which suggests its availability to replace the currently applied subjective pilling evaluation.