数字图像处理在织物疵点检测上的应用研究进展

Research Progress on the Application of Digital Image Processing in Fabric Defect Detection

[研究意义]长期以来,织物疵点检测过程仍通过人工目测来进行,成本高,检测率低,研究数字图像处理在织物疵点的运用,对纺织品质量控制具有重要的现实意义。[研究方法和内容]文章通过近年织物疵点检测,数字图像处理,人工神经网络等相关文献分析,简要阐述了近年来织物疵点在图像处理领域的检测方式,综述了基于共生矩阵和数学形态学的统计学方法,基于傅里叶变换,小波变换和Gabor的频谱方法,基于神经网络等的学习方法这三类织物疵点检测方法,对比总结了这些方法的历史发展和优缺点分布,并提出了自己的见解与展望,发现基于神经网络的学习方法识别率高,适用范围广,可以完成图像分类、语义分割等复杂的任务,加之其学习能力和容错特性,适用于现代企业对于不同织物疵点检测的要求。

For a long time, the fabric defect detection process is still carried out by manual visual inspection, which is of high cost and low detection rate. Research on the application of digital image processing in fabric defects has important practical significance in textile quality control. Through fabric defect detection, digital image processing, artificial neural network and other related literature analysis in recent years, this paper briefly expounded the detection methods of fabric defects in the field of image processing in recent years, reviewed statistical methods based on symbiotic matrices and mathematical morphology.Based on Fourier transform and the three types of fabric defect detection methods, such as wavelet transform and Gabor’s spectrum method, and learning method based on neural network, the historical development and distribution of advantages and disadvantages of these methods are compared and summarized, and their own insights and prospects are put forward. It is found that the learning method based on neural network has high recognition rate and wide scope of application, can complete complex tasks such as image classification and semantic segmentation, coupled with its learning ability and fault tolerance characteristics, it is suitable for the requirements of modern enterprises for different fabric defect detection.