Abstract: The structural parameters of woven fabrics mainly include warp and weft densities and fabric weave. The identification and analysis of these structural parameters is an important prerequisite for textile enterprises to conduct sample design, large-scale production, and quality control. At present, the traditional texture analysis of textile manufacturing mainly relies on external tools and artificial vision, which has strong subjective factors and low efficiency. With the continuous development of digital image processing technology, the traditional textile industry is gradually transitioning towards intelligence and automation. Research on detecting woven fabric weave parameters is more inclined towards automatic recognition. As for automatic recognition, cameras and machine vision are used to replace artificial vision for parameter detection of woven fabric images, which has good objectivity and improves the recognition efficiency, and woven fabric tissue information is quickly obtained. Therefore, automated identification of woven fabric parameters has become a research hotspot. There has been a lot of research on automatic recognition of woven fabric weaves nowadays, but there are still shortcomings in the segmentation and recognition of weave points. The segmentation of weave points only has high accuracy for the segmentation of ideal fabric images, while the recognition of weave points fluctuates greatly for different fabrics and different fabric images, indicating that the adaptability is not ideal.
In order to improve the accuracy and stability of recognition, the effect of Gaussian Mixture Model-EM (GMM-EM) algorithm on woven fabric weave recognition was studied. First, different fabric images were preprocessed and skew correctly to improve the subsequent segmentation of the weave points. Then, the grayscale projection method was used to locate the fabric weave points, and the grayscale co-occurrence moments of the tissue points were extracted as texture features. The texture feature data were dimensionally reduced by using principal component analysis. Finally, two kinds of common unsupervised learning were compared with the EM algorithm for Gaussian Mixture Models (EM-GMM), and four evaluation indicators of unsupervised learning were used for evaluation.
On this basis, for fabric images with warp and weft yarns not perpendicular to each other, the improved grayscale projection method in this paper was used to achieve fabric weave point positioning. The identification algorithm used in this paper was based on the Gaussian Mixture Model-EM (GMM-EM) algorithm to identify the weave chart.
The results of the four evaluation criteria indicate that the recognition algorithm proposed in this paper has high recognition accuracy and good adaptability, achieving automatic recognition of organizational points and outputting organizational charts. Compared with that of the other two clustering algorithms, the recognition performance of this algorithm has been effectively improved.

Key words: organization point segmentation, automatic identification, K-means clustering, FCM, Gaussian Mixture Models

摘要： 针对现有无监督学习识别机织物组织点的准确率相对较低和不稳定的问题，研究基于EM算法的高斯混合模型对机织物组织点的识别方法。首先对采集的不同织物图像进行预处理及图像矫正，以提高后续的组织点的分割效率；接着利用改进的灰度投影法进行织物组织点定位，并提取组织点的灰度共生矩作为纹理特征，通过主成分分析对纹理特征进行降维处理；最后采用2种常见无监督学习与文章所用的识别方法做实验比较，并采用4种评估指标进行评估，得到评估结果。通过计算4种评估指标平均值和标准差进行比较，文章所用识别方法的评估参数平均值都要比其余两种识别算法高。文章所用识别方法能对织物组织点进行自动识别，并且识别的准确率相比于其余两种识别算法得到了有效地提升。

关键词: 组织点分割, 自动识别, K均值聚类, 模糊C均值算法, 高斯混合模型