玻璃管缺陷检测系统及其性能优化

发布时间：2018-01-14 00:30

本文关键词：玻璃管缺陷检测系统及其性能优化　出处：《中国科学院重庆绿色智能技术研究院》2016年硕士论文　论文类型：学位论文

【摘要】：目前,工业中针对玻璃管是否存在缺陷的检测已经成为了生产过程中一个必不可少的环节,同时检测的要求也越来越高,即需要在保证精度、控制成本的基础上实现玻璃管缺陷的快速检测。对此,本文提出了一种基于GPU(Graphics Processing Unit)实现缺陷检测的方法,既可以保证检测的精度,成本也不会增加,同时又能极大提高系统的检测速度。该方法主要通过在CUDA(Compute Unified Device Architecture)架构下对图像处理过程中的形态学滤波(腐蚀和膨胀),边缘检测(高斯滤波、梯度和方向角计算、非极大值抑制、滞后阈值连接边缘)以及阈值化处理(Otsu二值化)三个过程进行改进,使算法运行时间极大减少。根据实验数据分析可得,GPU下进行一次完整检测所需时间远远小于CPU进行一次完整检测的时间,GPU下的整体运行效率较CPU下有约16倍的性能提升,同时检测精度相差无异。其中,通过对各个算法单次运行在GPU和CPU下的执行效率进行对比,可以得出GPU下形态学滤波有近14倍的提升,边缘检测有约15倍的提升,Otsu阈值化有近9倍的提升。因此,通过GPU实现系统检测算法的并行化能极大减少算法的运行时间,提升检测系统的计算性能。
[Abstract]:At present, the detection of glass tube defects in industry has become an essential link in the production process, and the requirements of the detection are increasingly high, that is, the need to ensure the accuracy. On the basis of cost control, rapid detection of glass tube defects is realized. This paper presents a method of defect detection based on GPU(Graphics Processing Unit, which can not only guarantee the accuracy of detection, but also do not increase the cost. At the same time, it can greatly improve the detection speed of the system. Morphological filtering (corrosion and expansion) in image processing under the framework. Edge detection (Gaussian filtering, calculation of gradient and direction angle, non-maximum suppression, lag threshold linking edge) and threshold processing Otsu binarization are improved. According to the analysis of the experimental data, it can be found that the time required for a complete detection under CPU is much less than that for a complete detection by CPU. The overall operating efficiency of GPU is about 16 times higher than that of CPU, and the detection accuracy is the same. By comparing the execution efficiency of each algorithm under GPU and CPU, it can be concluded that the improvement of morphological filtering and edge detection under GPU is nearly 14 times and 15 times higher respectively. Therefore, the parallelization of the system detection algorithm through GPU can greatly reduce the running time of the algorithm and improve the computing performance of the detection system.
【学位授予单位】：中国科学院重庆绿色智能技术研究院
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TQ171.65;TP391.41

【相似文献】