Abstract:

Objective The performance of yarn directly determines the quality of the fabric, and the unevenness of yarn is an important indicator for evaluating yarn quality. Testing and analyzing yarn unevenness is a necessary aspect for controlling and improving yarn quality. However, as a typical linear heterogeneous material, yarn is difficult to accurately characterize its actual performance by simply detecting its volume or single direction projection using existing detection instruments. In order to more accurately represent the three-dimensional features of yarn, a method based on multi-view yarn images is proposed to measure and evaluate the yarn unevenness.

Method An image acquisition device with single camera was constructed to obtain images of yarn from multiple view angles, before the obtained yarn images in four directions were processed sequentially through automatic threshold segmentation, image automatic cropping, rapid removal of hairiness, and removal of isolated areas. A clear and noise-free image of the yarn core was obtained. Single- and multi-view diameters and CV values were read from the yarn core image, and new multi-view unevenness mean (CV_n) indicators and three-dimensional unevenness variation coefficient (S) indicators were proposed based on the elliptical model to characterize yarn unevenness.

Results The unevenness of five cotton ring spun yarns with different linear densities was tested and compared with the test results of Uster tester. The results showed that the difference in yarn diameter detected by the two testing methods is not significant, both illustrating diameter decreases as the linear density was decreased. The average diameters measured from four single perspectives was quite consistent, indirectly indicating that the diameter testing process of the proposed method is correct. The yarn diameter measured by the proposed method was slightly smaller than that measured by the Uster tester, mainly because of the use of image processing operations to remove hairiness. The results also showed that CV 0.3 mm, CV 8 mm under single and multiple view angles was consistent with the results measured by Uster tester. The CV values gradually increase with the decrease of linear density, which was consistent with the theory that says the smaller the linear density, the more uneven the yarn. The proposed three-dimensional unevenness variation coefficient S, regardless of S 1p, S 0.3 mm, or S 8 mm, increased with the decline of yarn linear density, and all conform to the theory that the larger the fiber length, the smaller the CV value. The proposed method approximated the yarn with a regular ellipse for cross-sectional area, and the diameter values of multiple perspectives were multiplied using the ellipse area formula. As the area increases compared to the diameter value, the CV value also increases. Therefore, the measured S values are greater than the Uster tester results at various cutting lengths, thus can characterize the variation rule of yarn thickness.

Conclusion In this paper, a new image processing method and a new indicator for characterizing yarn unevenness are proposed based on a multi-image acquisition device with single camera. The yarn detection results are compared with the diameter and CV values measured by the Uster tester and the proposed method. The results show that in terms of yarn appearance diameter, single-view and multi-view CV, the two methods lead similar results with consistency, which are in line with relevant theory and understanding, suggesting that the method proposed in this paper could be used for detecting yarn unevenness. However, this paper used a rotating camera to capture multi-view images, which caused inefficient. What's more, the yarn cross-sectional area is regarded as a regular ellipse to calculation. After the diameter values of multiple angles are multiplied according to the ellipse area formula, the data base value becomes larger, and then the CV value becomes larger. This is also the main reason why S is too large in this paper. In the future, a multi-camera real-time yarn detection device should be built up to facilitate three-dimensional real-time yarn detection.

Key words: yarn quality, yarn unevenness, multi-view image, image segmentation, elliptic model, three-dimensional detection