您好, 访客   登录/注册

基于并行综合学习粒子群算法的织物图像疵点检测

来源:用户上传      作者:葛芸萍

   摘 要:为了提高织物图像疵点检测的质量,提出了并行综合学习粒子群算法。首先,通过织物透光率获得织物图像的疵点;接着多尺度利用织物图像灰度值差异对疵点区域显著性增强,把疵点与周围像素进行区分,从而弱化背景对织物疵点的影响;然后综合学习粒子增设局部吸引因子,多群和并行策略提高搜索能力;最后得出算法流程。实验仿真显示本文算法对疵点检测清晰,破损疵点检测准确率为88.15%,缺失疵点检测准确率为90.46%,移位疵点检测准确率为93.87%,断经疵点检测准确率为86.54%,高于其它算法,同时检测消耗时间较少。
  关键词:织物;疵点;综合学习;并行;粒子群算法;检测
  中图分类号:TP393
   文献标志码:A
   文章编号:1009-265X(2022)04-0142-07
  Fabric image defect detection based on parallel comprehensivelearning particle swarm optimization
  GE Yunping
  (Yellow River Conservancy Technical Institute, Kaifeng 475004, China)
  Abstract: In order to enhance the quality of fabric defect image detection, a parallel comprehensive learning particle swarm optimization algorithm is proposed. Firstly, the fabric defect in the image was obtained according to the light transmittance of the fabric. Secondly, through multi-scale utilization of gray value difference of fabric image, the saliency of the defect area was enhanced, and the defect was distinguished from the surrounding pixels, thereby weakening the influence of the background on the fabric defect. Thirdly, the local attraction factor was added based on comprehensive learning particles, and the multi-swarm and parallel strategies were adopted to improve the search capability. Finally, the algorithm flow was obtained. The simulation results show that the algorithm proposed in this paper can clearly detect the defect. The detection accuracy of damaged defect of 88.15%, the detection accuracy of missing defect of 90.46%, the detection accuracy of shifted defect of 93.87%, and the detection accuracy of broken warp defect of 86.54%. Its detection accuracy is higher than other algorithms, and consumes less time than other algorithms.
  Key words: fabric; defect; comprehensive learning; parallel; particle swarm optimization algorithm; detection
  织品在生产过程中受原材料、静电、温湿度等因素影响,会产生各种各样的疵点并严重影响纺织品的使用价值,因此在验布过程中需要及时检测出疵点[1]。人工方法对织物疵点检测是在无眩光的背面窗旁或日光灯照明条件下通过经验检测。此方法简单,但是存在效率低、漏检率高等缺点,检验结果正确率较低。自动化检测织物疵点已成为对织物质量进行控制和实现织造及验布工序的关键环节,因此受到了纺织行业的关注。目前较为成熟的织物图像疵点检测算法有:小波变换(Wavelet transform,WT)检测方法[2],如果边缘周围存在与疵点相似的痕迹,就很容易造成算法的误判,对疵点定位效果明显下降。均值滤波(Mean filtering,MF)检测方法计算比较简单[3],但是不能很好地保护图像细节,对图像边缘的处理效果比较差。神经网络(Neural network,NN)算法[4],训练神经网络需要大量的数据样本,但是织物生产过程的复杂性,随机出现的疵点很可能存在与训练模型不一致,难以准确判断疵点,同时神经网络参数值难于确定。裂变粒子滤波(Fission particle filter,FPF)算法[5],粒子的多次裂变生成更多的粒子,粒子滤波获得疵点区域的最佳分割阈值,但是粒子多次裂变的次数不易控制。双混沌机制粒子群(Double chaotic mechanism particle swarm optimization,DCMPSO)算法[6],在粒子群的全局、局部寻优中采用不同的混沌机制,利于粒子群搜索到最优解,但是双混沌机制增加了算法的复杂性,检测速度较低。改进宇宙(Improved universe,IU)算法[7],通过设计宇宙空间拓扑结构以及宇宙进化策略使得检测率上具有较大的优越性及良好的适应性,但是宇宙空间拓扑结构对宇宙之间移民、交互存在制约性,导致具有倾斜度的疵点无法检测出。

nlc202209071010



转载注明来源:https://www.xzbu.com/1/view-15439271.htm

相关文章