【作者】 姜颖军；

【导师】 陈元琰；

【作者基本信息】 广西师范大学 ， 无线电电子学， 2000， 硕士

【摘要】 在玻璃制品的生产过程中，随着生产速度的提高以及越来越严格的质量要求，传统的人工检测裂纹的方法由于受主观因素的影响，已不能保质保量的完成生产任务。在这种情况下，国内外不少厂家开始研制用于玻璃制品裂纹检测的机器。从已开发的产品来看，它们都具有机械制造要求高、不易调整且价格昂贵等缺点。 随着计算机软件，硬件的发展，以及计算机视觉相关理论的完善，用计算机视觉进行裂纹检测的方法已变得切实可行。在国外已有少数厂家利用这一原理制造出较为成功的产品，但在国内还处于起步阶段。按此方法制造的检测设备相对以往的采用机械接触方式进行检测的设备，它具有以下优点： 制造代价小，对机械制造的精度要求低； 灵活性高，易于调试； 系统的检测速度可随CPU速度的提高而不断增加； 工作环境要求低。 作者研制的“玻璃瓶裂纹检测系统”是“玻璃制品的计算机视觉在线检测系统”的一个子系统。整个系统由广西师范大学电子研究所与桂林市玻璃厂协作开发。对于用计算机视觉进行玻璃制品裂纹的检测，该系统提出了在国内尚属全新的两种方法：简易判断法和边界特征判断法。这两种方法都采用非接触的传感方式。首先，选取合适的相机、光源、图像卡，在特定的空间位置获取能反映裂纹存在的一系列数字图像。由于是利用裂纹的反射光来进行判断，而裂纹在玻璃瓶上的位置并不确定，玻璃瓶的四周都有存在的可能性，因此必需对玻璃瓶进行旋转拍摄。系统采用的方法是固定相机和光源，旋转玻璃瓶，对每个玻璃瓶以固定频率摄取一系列图像，然后直接在用户内存对图像进行各种预处理，在预处理的基础上对图像进行分割，以便把裂纹小区和其他干扰小区分割开来，再对各小区提取特征，最后，根据各特征参数，使用不同的判断法则进行判断。 简易判断法和边界特征判断法采用相同的途径获取图像。在图像预处理阶段，前者使用邻域平均法进行噪声平滑，后者使用中值滤波法进行噪声平滑，这两种平滑方法对物体边缘产生不同的影响。邻域平均法使图像边界模糊，中值滤波法能保持图像边界基本不失真。在图像分割阶段，两者都使用门限化法(最佳阈值法)进行图像分割。在简易判断法中，抽取光斑面积和光斑位置两类特征作为判断依据。而在边界特征判断法中，在门限化的基础上用拉普拉斯算子进行边缘检测，提取各区域边缘，计算边界链码，然后再使用傅立叶级数提取边界特征细长度和圆形度作为判断依据。 在都能保证对裂纹进行有效判断的前提下，相对于简易判断法，边界特征判断法对用于检测的机械设备要求低，但是处理速度慢。其实质是牺牲部分处理速度来换取低价格、低精度的机械设备。 作者用Visual C++较成功地实现玻璃瓶裂纹的检测。更多还原

【Abstract】 In the production of glassware , with the enhancing of production speed and more and more strict request of quality , the traditional method detecting crack can not do well because of major factor. This method can not insure the quality and quantity of product. In this case , many factories begin to manufacture the machine that can detects the crack of glassware . The product those have been developed have some default , such as highly precision with machine manufacturing, uneasy adjusting and expensive price .With the development of computer software and hardware and perfecting of computer vision’s theory .In foreign countries some factories has made out this kind of product. Its quality is good. Our country is studying this kind of product and has not make out any kind of finished product .The method for detecting crack with computer is becoming viable . Compare to the traditional product that is made with the way of mechanical touching, The machine according to this kind of method to product has virtues:Low manufacturing price and low precision to mechanical manufacturing ;High feasibility and easy adjusting ;The detecting speed of system can enhance with the increase of CPU speed.Low request to working environment.Crack Detecting System of glass bottles is a subsystem of on-line detecting system of glassware with computer envision . We cooperate with Guilin Glass Factory to develop this system . I have invented two new way in this kind of system . Easy judging method (EJM) and border feature judging method (BFJM).These ways all follow the principle of computer vision .First , choosing right camera , lighting and image card to get a series of digital image with special space that can show crack . System uses the reflect of crack to judge, and the crack location of bottle is not static. The crack may lies in each place of a bottle. In this system when a series of digital images are being gotten, the camera and the light source are static, but the glass bottle is swirling .The Second , this system does some kinds of pre-processing in user buffer memory . After pre-processing , in order to distinct the crack area (CA) and disturbing area (DA) system segment the image . Then system extracts features from each small area . Last I judge the glass bottle in judging method according to all kinds of parameters .EJM and BFJM both choose a same way to get digital image . In the phase of image preprocessing ,the former uses adjacent area average method (AAAM) to smooth noise , the latter uses median fitting method (MFM) to smooth noise .These two method smoothing effect the border of area differently .AAAM faints the border of a image . MFM can maintain the border of a Image not to distort .In the phase of segmenting a image ,EJM and BFJM both use the best threshold method (BTM) to segment images . In the EJM , facula area feature and facula location feature are extract to judge bottles . In the BFJM , Laplacian is used to detect the border basing BTM .then 1 extract the border of each area and calculate the chain code of the border . afterwardsthe thin degree and circle degree with Fourier are extracted to judge bottle .In the precondition of judging crack rightly .comparing to EJM , BFJM has low request to machine for detecting . but its processing speed is slow . The essence is sacrificing part of processing speed to exchange the low price and low precision .This system realizes detecting the crack of bottles with Visual C++ .更多还原