Carpet Back Sizing Quality Assessment by Measuring the Amount of Resin Using Image Processing and Machine Learning Approaches

Abstract

The mechanical properties of the carpet, such as dimensional stability, bending stiffness, handle and creeping on the surface during use, have a direct relationship with the amount of resin applied to the back of the carpet in the sizing process. In today’s factories, the optimal amount of resin and the mechanical quality of the carpet are controlled by the operator touching the carpet on the machine carpet finishing line or manually while rolling the carpet. Proposed in this paper is an automatic method based on the evaluation of the bending stiffness of the sized carpet that uses digital image processing and machine learning to measure the optimal amount of size concentration and control this index. For this purpose, during the final stage of carpet production, the carpet is folded in the middle, and two edges of the carpet are placed on top of each other. A side view image is then taken of the carpet. Using edge detection methods, the edges of the carpet are identified, and different features, such as the average, maximum and minimum statistics for the curve and contour angles, are then extracted. Different conventional machine learning approaches, such as KNN, CART and SVM, are applied. To evaluate the proposed method, a dataset containing 220 different images is used in a 10-fold cross-validation scheme. Different performance measures resulting from the evaluations demonstrate the effectiveness and applicability of the method.

Keywords

• carpet quality assessment
• carpet back sizing
• digital image processing
• machine learning
• edge detection