基于机器视觉的芯棒缺陷检测与分类算法研究

发布时间：2018-06-02 10:35

本文选题：复合绝缘子芯棒 + 缺陷分割　；参考：《北京邮电大学》2016年硕士论文

【摘要】：基于视觉的缺陷检测技术以其高准确度、高效率等优势在工业检测领域被广泛应用。复合绝缘子芯棒质量是影响绝缘子质量的重要因素,芯棒缺陷的检测比较困难。传统人工使用肉眼进行检测的方法存在可靠性差,效率低下的等缺陷。将基于机器视觉的缺陷检测技术应用于绝缘子芯棒自动缺陷检测与分类,对于提高芯棒质量检测的效率和可靠性具有重要意义。论文研究基于视觉的芯棒缺陷检测与分类算法并研制芯棒缺陷检测软件。论文针对芯棒的特点展开一系列算法研究。论文的主要工作如下:(1)针对芯棒缺陷的特点,研究了基于视觉显著性模型的缺陷目标定位与分割算法,取得了良好的缺陷定位与分割效果。(2)基于不同的芯棒样本,对芯棒缺陷区域的特征进行了分析,研究能够标识芯棒缺陷的特征,包括缺陷区域的长宽比、方向、形心位置、矩形度、致密度、平均灰度值、灰度均方差。(3)基于BP神经网络对缺陷图像进行分类。分析并确定了神经网络模型参数的选择和训练过程。(4)设计了芯棒缺陷检测软件系统的方案,并完成了缺陷检测软件系统的开发。
[Abstract]:Vision-based defect detection technology is widely used in the field of industrial detection because of its high accuracy and high efficiency. The quality of composite insulator mandrel is an important factor affecting the quality of insulator, and the detection of mandrel defects is difficult. The traditional manual inspection method with naked eye has some defects, such as low reliability and low efficiency. The application of machine vision based defect detection technology to the automatic defect detection and classification of insulator mandrel is of great significance for improving the efficiency and reliability of mandrel quality detection. In this paper, visual-based mandrel defect detection and classification algorithms are studied and a mandrel defect detection software is developed. In this paper, a series of algorithms are studied according to the characteristics of mandrel. The main work of this paper is as follows: (1) aiming at the characteristics of mandrel defects, a defect target localization and segmentation algorithm based on visual salience model is studied. A good defect localization and segmentation effect is obtained. The characteristics of mandrel defect region are analyzed, and the characteristics that can identify the mandrel defect are studied, including the aspect ratio, direction, centroid position, rectangle degree, density, average gray value, etc. Based on BP neural network, defect images are classified. The selection and training process of neural network model parameters are analyzed and determined. The scheme of mandrel defect detection software system is designed, and the development of defect detection software system is completed.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】