可重构的机器视觉在线检测方法的研究

发布时间：2018-09-09 20:03

【摘要】：质量需求是现代企业的核心需求，产品质量的好坏直接关系着企业的市场竞争力和企业自身的存续。因此，将质量放在首位是目前所有企业的共同选择。质量控制与质量管理的方法与手段多种多样，传统检测与测量往往依赖于人工完成。本文着眼于流水线上、不间断运动的连续物体视觉可识别量的在线检测。利用机器视觉替代人工检测，运用计算机及软件替代检测工人，通过可重构的视觉检测方法实现在不同被检对象上的快速反应与装备实现。本文主要研究了可动态增减检测子系统的分布式网络拓扑，支撑采集控制硬件层、图像处理识别层、界面显示层、质量等级评价等多层面的视觉重构理论，建立了基于软件芯片的视觉重构体系，对重构流程、重构方案进行了设计，采用规范化、标准化的方式设计了各个层次间交互接口。针对市场上不同厂商的数字图像获取设备，本文均根据自身产品而定制、通用性差的缺点进行了相关研究。首先，分析了不同数字图像获取标准的优劣，建立了异构硬件环境下的图像获取通用模型；接着，创建了通用图像获取接口，设计并完善了SDK，包括初始化函数、设置函数、获取函数、图像处理回调函数、存储函数、辅助函数在内的六大类函数的定义；最后，对函数接口及其组态设计进行了研究。在图像获取的基础上，对图像处理环节进行了探讨。首先，分析了视觉检测的基本需求，梳理了视觉检测的基本流程；接着，根据视觉检测流程的主要需求，设计了机器视觉在线检测算法库，将常用的图像预处理、图像分割、图像复原、特征提取等算法进行了标准化、参数化设计，着重突出优化算法代码、执行效率、鲁棒性等问题；最后，建立并完善了一套基于配置信息的视觉检测运行环境，可实现检测算子的搜索、调用、重载等功能，，有利于提高检测系统的柔性，加速针对不同检测对象的设备部署。为实现图像特征的提取与识别，本文首先研究了常用特征描述方法，分析了其优缺点，并指出单一特征或过小规模特征集在可重构视觉检测方法中的局限性。其后，针对这一核心问题展开，设计了一套基于统计与时频联合的特征提取方法，建立了通用性机器视觉在线检测特征集，并利用该特征集对三种产品的图像图形学特征进行了描述。最后，尝试设计了基于遗传算法的特征解耦方法，并就其中关键技术进行了研究。最后，利用可重构的机器视觉方法对粘扣带、导爆管、网孔织物外观质量视觉检测进行了研究，开发了具有自主知识产权的检测系统，验证了可重构的产品视觉检测方法的可行性和有效性。
[Abstract]:Quality demand is the core demand of modern enterprises. The quality of products is directly related to the market competitiveness of enterprises and the survival of enterprises themselves. Therefore, quality in the first place is the common choice of all enterprises. There are a variety of methods and means for quality control and quality management, and traditional inspection and measurement often rely on manual completion. This paper focuses on the on-line detection of visual recognizable quantities of continuous objects with continuous motion on pipeline. Machine vision is used to replace manual detection, computer and software are used to replace workers, and reconfigurable visual detection method is used to realize rapid response and equipment on different objects. This paper mainly studies the distributed network topology of dynamic incremental and subtractive detection subsystem, supports the collection and control hardware layer, the image processing recognition layer, the interface display layer, the quality grade evaluation and so on multi-layer vision reconstruction theory. The visual reconfiguration system based on software chip is established. The reconstruction process and reconfiguration scheme are designed. The interface between different levels is designed in a standardized and standardized way. Aiming at the digital image acquisition equipment of different manufacturers in the market, this paper studies the shortcomings of customization and poor generality according to their own products. Firstly, the advantages and disadvantages of different digital image acquisition standards are analyzed, and the general image acquisition model under heterogeneous hardware environment is established. Then, the general image acquisition interface is created, and the initialization function and setting function are designed and perfected in SDK,. Get function, image processing callback function, storage function, auxiliary function, etc. Finally, the function interface and its configuration design are studied. On the basis of image acquisition, the link of image processing is discussed. Firstly, the basic requirements of visual detection are analyzed, and the basic flow of visual detection is combed. Then, according to the main requirements of visual detection process, the online detection algorithm library of machine vision is designed to preprocess and segment the commonly used images. Image restoration, feature extraction and other algorithms are standardized, parameterized design, focusing on the optimization of algorithm code, execution efficiency, robustness and other issues. Finally, a set of visual detection environment based on configuration information is established and improved. The functions of searching, calling and overloading of detection operators can be realized, which can improve the flexibility of the detection system and accelerate the equipment deployment for different detection objects. In order to achieve image feature extraction and recognition, this paper first studies common feature description methods, analyzes their advantages and disadvantages, and points out the limitations of single feature or too small feature set in reconfigurable vision detection method. Then, aiming at this core problem, a set of feature extraction method based on statistical and time-frequency combination is designed, and the feature set of universal machine vision on-line detection is established. The feature set is used to describe the image graphics features of three kinds of products. Finally, the feature decoupling method based on genetic algorithm is designed and the key technologies are studied. Finally, using the reconfigurable machine vision method to study the visual inspection of the appearance quality of the adhesive tape, the detonation tube and the mesh fabric, the inspection system with independent intellectual property rights is developed. The feasibility and effectiveness of the reconfigurable visual inspection method are verified.
【学位授予单位】：武汉科技大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TP274;TP391.41

【参考文献】