虚拟样本在粗糙度视觉测量中的应用方法研究

A study of the application of virtual samples in visual roughness measurement

在基于机器视觉的粗糙度测量系统中,最为关键的是建比图像信息与粗糙度关联指标的预测模型,充足的样本量是建立有效预测模型的基础。针对以上问题,文章提出了虚拟样本在粗糙度视觉测量中的应用方法,对通过图像采集系统与粗糙度测量仪获得对应粗糙度值的原始样本,分别采用重采样、奇异值重构、正三角重构加权融合的虚拟样本生成方法扩充样本量,基于灰度共生矩阵提取样本纹理特征,结合神经网络建空并训练图像信息与粗糙度关联指标的预测模型,结果表明:扩充样本量可有效提高粗糙度视觉测量的准确率,奇异值重构法提高了10.3%、重采样法提高了12%、正三角重构加权融合法提高了16.5%,证明了虚拟样本生成方法应用于粗糙度视觉测量中的可行性,为粗糙度的在机检测提供理论基础。

In the roughness measurement system based on machine vision,the most important thing is to establish a prediction mode of correlation index between image information and roughness,while a sufficient sample size is the foundation for setting up such an effective prediction mode.In view of the above-mentioned problems,this paper proposes an application of virtual samples in roughness visual measurement.For the original samples with corresponding roughness values obtained through image acquisition system and roughness measuring instrument,the virtual sample generation method of weighted-fusion-resampling,singular value reconstruction and trigonometric reconstruction is used to expand the sample size,extract the sample texture features based on the grey level co-occurrence matrix,and establish and train the prediction model of image information and roughness related indicators combined with neural network.The experimental results show that the expansion of sample size can effectively improve the accuracy of roughness visual measurement,the singular value reconstruction method improves by 10.3%,the resampling method improves by 12%,while the regular triangle reconstruction weighted fusion method by 16.5%.The feasibility of applying virtual sample generation method to roughness visual measurement provides a theoretical basis for on-machine roughness measurement.