基于不变矩与神经网络的织物缺陷检测研究  

Research of Fabric Defect Detection Based on Invariant Moments and Neural Network

作　　者：逄鹏[1] 向洪波[1] 魏喜雯[1] 杨新年[1] 

机构地区：[1]黑龙江工业学院电气与信息工程系,黑龙江鸡西158100

出　　处：《针织工业》2015年第11期69-72,共4页Knitting Industries

摘　　要：针对织物表面的非结构化畸变缺陷,提出一种基于不变矩与神经网络相结合的织物缺陷检测方法。首先,利用图像处理的方法对织物图像进行预处理分块;其次,提取待检测织物分块图像的7个不变矩值,通过比较分块变换后的统计特征量,发现缺陷织物与非缺陷织物的统计特征量存在差异;基于该特点,建立BP神经网络,应用织物图像的7个不变矩特征值作为神经网络的输入,通过学习大量样本,获取最佳权值参数,实现对非结构化畸变缺陷的识别分类。试验结果表明,该算法检测的准确率达到80%以上,对破洞的检出率接近100%,能够很好检测织物的非结构化缺陷,有效地满足织物的生产工艺要求。In order to defect the unstructured distortion defects on the fabric surface, a defect detection method based on the invariant moment and neural network was presented. Firstly, the fabric image is predivided into blocks with the method of image processing; Secondly, the seven moment invariants of the image from the fabric blocks .are extracted, by comparing the statistical feature of the block transformation, there are differences in statistical characteristics between the defects and the non-defect fabrics; On the basis of the above result, the BP neural network was established, the seven moment invariants of fabric image were selected as the input of neural network, after all neural network weight parameters were adjusted to the optimization through sample training, realize the recognition and classification of unstructured distortion defects. The results show that the inspecting accuracy, detection of the broken holes of the algorithm are above 80% and approximately 100%, and it can detect the unstructured distortion defects on the fabric surface very well and satisfy the requirement of the fabric production process effectively.

关 键 词：织物缺陷检测 不变矩 神经网络 分类识别 

分 类 号：TS107[轻工技术与工程—纺织工程]