Intelligent Fabric Hand Prediction System With Fuzzy Neural Network

Fabric selection is a crucial step in fashion product development. Prior research works have studied the prediction of fabric specimens based on the fabric hand descriptors via either traditional statistical methods or artificial intelligence methods. Despite showing good prediction accuracy, these methods usually lack an understandable ruleset, which means their “interpretability” is low. In this paper, a fuzzy neural network (FNN) based intelligent fabric hand prediction system is explored. Unlike some traditional FNN models in which a full ruleset of the artificial neural network (ANN) is presumed, the proposed FNN system includes a simplification of the network structure and feature selection, so that the number of rules is significantly reduced without big sacrifice on prediction accuracy. Real datasets collected from 30 participants' evaluation on a set of ten fabric specimens are used to train and test the performance of the proposed system. The system's prediction accuracy is found to be over 80%. Applications of the proposed system are discussed and future research directions are outlined.

I. Introduction

Materials and fabrics used in the textile industry give various “touch feelings.” These feelings are closely related to the consumer needs from the market. The fabric hand, which describes the touch feeling of a piece of fabric, and is hence, related to the comfort, style, and appearance of textile, is an essential sales element of a garment. Researchers have attempted to evaluate fabric hand in a systematic way, since the 1950s [1]. Essentially, fabric evaluation is performed in two main ways: subjective evaluation and objective evaluation. Individual consumers prefer to use their sensory feelings (subjective evaluation) instead of the mechanical properties to evaluate the fabric performance. Obviously, subjective evaluation is based on human sensitivity and experience, and hence, it varies from person to person. In the literature [10], there are proposals on how the sensory descriptors could be organized for better communications of the sensory meanings and Table I shows the 14 fabric hand descriptors with their respective English descriptions as explored in prior studies [8] [11] [12]. Table I Fourteen Bipolar Fabric Hand Descriptors with Their English Description

Descriptor index	Bipolar adjective pairs	English word description
S1	Flat H textured	Smooth and level	Having a certain kind of texture
S2	Not bulky ↔ bulky	Slim size, shape, mass or quantity	Great size, shape, mass or quantity
S3	Light ↔ heavy	Easy to see in, or bright	Having a certain weight
S4	Thin ↔ thick	Having a small distance between opposite surface	Having a large distance between opposite surface
S5	Silky ↔ scratchy	Like silk, soft, smooth or shiny	Spoilt by scratches
S6	Smoothness ↔ roughness	Having an even surface, or not rough	Having an uneven surface, or not smooth
S7	Fine ↔ coarse	Very thin	Neither fine nor smooth, Rough
S8	Limp ↔ crisp	Lacking strength or stiffness	Firm, fresh
S9	Flexible ↔ stiff	Could be bent easily	Not easily bent
S10	Soft ↔ hard	Not firm against pressure	Firm and stiff, not easily broken or bent
S11	Flimsy ↔ firm	Strong, solid, or hard	Not strong, light or thin
S12	Compact ↔ loose	Pressed, joined together or united firmly and closely	Not fastened, tied up or shut up
S13	High drape ↔ low drape	Easy to hang or stretch out loosely and carelessly	Difficult to hang or stretch out loosely and carelessly
S14	Cool ↔ warm	Neither warm nor cold. Pleasantly cold	Having a pleasant feeling of heat

References and Sources: [8], [11], [12].