基于视网膜感知机制和机器学习的工业视觉检测理论研究

项目名称： 基于视网膜感知机制和机器学习的工业视觉检测理论研究

项目编号： No.51305214

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 机械、仪表工业

项目作者： 梁冬泰

作者单位： 宁波大学

项目金额： 23万元

中文摘要： 借鉴人眼的敏锐性能和人的学习能力，本项目基于视网膜感知机制和机器学习方法实现一种通用的工业视觉检测理论；研究模拟视网膜感知机制的图像特征表示和提取方法，针对常见的工业视觉检测应用，提取图像中的能量特征、形状特征、空间分布特征、颜色特征、纹理特征等几种典型的潜在分类特征，以及这些特征所符合的线性的、非线性的、基于核函数、马尔可夫模型等几种典型的统计分析模型；以分类器最佳检测性能为优化目标，利用机器学习方法从特定检测任务的训练样本中，自动选择一个或多个图像特征表示，自动选择合适的统计分析模型，完成该检测任务的最优分类器设计和实现；结合表面缺陷检测和颜色纹理分级的具体应用开展实验研究，测试分类器的检测性能，验证工业视觉检测理论方法的可靠性；本项目从方法论的角度建立工业视觉检测理论的基本框架，为各类工业视觉检测系统的设计和优化提供理论指导，为解决复杂的工业视觉检测问题提供参考。

中文关键词： 工业视觉检测；图像特征表示；特征选择机制；图像分类器；机器学习

英文摘要： Inspired by the visual acuity of the eyes and the learning ability of human, this proposal presents an industrial visual inspection theory based on the retina perception mechanism and machine learning. The research on the method of image feature representation and extraction is carried out by simulating the retina perception mechanism. For the several common visual inspection applications, a few of typical potential classification features, such as energy features, geometrical features, spatial features, color and texture features, are extracted from the application images with the typical statistical models which these features could fit. These statistical models such as linear, non-linear, kernel-based, Markov models are investigated in the research. In order to optimize the performance of classifier for visual inspection, the machine learning methods are used to select one or more optimal image features automatically from those above typical potential classification features based on the training image database of the specified inspection tasks. And the suitable statistical models for the optimal features are also selected automatically according to the optimization criteria. So the design and implementation of optimal classifier for the specified inspection tasks are achieved by machine learning method. In o

英文关键词： industrial visual inspection；image feature representation；feature selection mechanism；image classifier；machine learning