作者：岑若晨, 李建良

作者单位：南京理工大学理学院, 江苏 南京 210094

关键词：连铸坯缺陷图像;二维Otsu法;粒子群优化;动态惯性权重;变异

摘要：

为更准确有效地提取连铸钢板坯图像中的各类缺陷，通过改进标准PSO算法以优化二维Otsu的阈值选取。将二维Otsu类间方差函数作为粒子的适应度函数，根据适应度值使粒子的惯性权重自适应地优化，根据迭代次数对粒子的变异概率进行改进，提高全局寻优能力和收敛精度。最后选取连铸坯不同种类的缺陷图像进行分割实验，对比二阶振荡PSO-Otsu、二维Otsu、SPSO-二维Otsu法和改进算法的分割结果，多次实验结果表明，改进算法对各类缺陷的分割准确率和成功率分别在90%和96%以上，且算法运行快，具有较好的实用性。

Application of improved PSO-2D Otsu method in continuous casting billet defect image segementation
CEN Ruochen, LI Jianliang
School of Science, Nanjing University of Science & Technology, Nanjing 210094, China
Abstract: In order to extract various defects in the continuous cast steel slab image more accurately and effectively, the standard PSO algorithm is improved to optimize the threshold selection of two-dimensional Otsu. The 2D Otsu inter-class variance function is used as a fitness function of the particles, and the inertia weight of the particles is adaptively optimized according to the fitness value, then the variation probability of the particles is improved according to the number of iterations, and the global optimization capability and the convergence precision are also improved. The results of the segmentation of the different kinds of defect images of the continuous casting blank are compared, and the results of the segmentation of the second-order oscillating PSO-Otsu, the 2D Otsu, the SPSO-2D Otsu method and the improved algorithm are compared. The experimental results show that the segmentation accuracy and success rate of the improved algorithm are above 90% and 96%. It runs fast, and has good practicability.
Keywords: continuous casting defect image;2D-Otsu;particle swarm optimization;dynamic inertia weight;mutation
2020, 46(4):19-24  收稿日期: 2019-06-17;收到修改稿日期: 2019-07-17
基金项目:
作者简介: 岑若晨(1994-),女,河南开封市人,硕士研究生,专业方向为计算技术及其应用软件研究
参考文献
[1] 葛小波. 连铸坯质量监视管理系统[J]. 连铸, 2019, 44(1):70-72
[2] 赵梦超, 孔令成, 谭治英. 基于改进Otsu法的镀膜金属带缺陷分割[J]. 计算机工程与设计, 2018, 39(9):2811-2816
[3] 化春键, 周海英. 阈值分解下的冷轧极薄带钢表面缺陷分割[J]. 机械科学与技术, 2017, 36(2):308-313
[4] KENNEDY J, EBERHART R. Particle swarm optimization[C]//IEEE International Conference on Neural Networks, 1995, 4:1942-1948.
[5] 王曼, 王姮, 张华. 基于简化PSO优化的高铁钢轨智能检测技术[J]. 信息技术与网络安全, 2018, 37(2):77-80
[6] SHI Y, EBERHART R, Empirical study of particle swarm optimization[M]. 2002, 1945-1950.
[7] 刘桂红, 赵亮, 孙劲光, 等. 一种改进PSO优化算法的Otsu图像阈值分割方法[J]. 计算机科学, 2016, 43(3):309-312
[8] LIN J, WU S. A PSO-based algorithm with subswarm using entropy and uniformity for image segmentation[C]//Genetic and Evolutionary Computing, 2012, 10:500-504.
[9] HELEN R, KAMARAJ N, SELVI K, RAJA V. Segmentation of pulmonary parenchyma in CT lung images based on 2D Otsu optimized by PSO[C]//Emerging Trends in Electrical and Computer Technology, 2011, 10:536-541.
[10] 李鑫, 崔昊杨, 霍思佳, 等. 基于PSO优化法的Niblack电力设备红外图像分割[J]. 红外技术, 2018, 40(8):780-785
[11] 曹爽, 安建成. 狼群优化的二维Otsu快速图像分割算法[J]. 计算机工程与科学, 2018, 40(7):1221-1226
[12] 彭启伟, 罗旺, 冯敏, 等. 改进二维Otsu法和果蝇算法结合的图像分割方法[J]. 计算机应用, 2017, 37(S2):193-197
[13] JIANG Y, YANG Z, HAO Z. A cooperative honey bee mating algorithm and its application in multi-threshold image segmentation[C]//Evolutionary Computation, 2014, 10:1579-1585.
[14] 连铸钢板坯低倍组织缺陷评级图:YB/T 4003-2016[S]. 冶金工业出版社, 2016.