【作者】 白伟；

【导师】 李红松；

【作者基本信息】 云南大学 ， 控制工程， 2017， 硕士

【摘要】 烟包外观的好坏直接影响产品质量,质量是企业的生命。要想在日趋激烈的卷烟市场保持一定的市场占有率,质量是关键。烟包外观工艺质量包括商标包装是否到位、破损、反包装、划痕,开胶等;税票裁切是否到位、粘贴位置规整与否、破损、折角等。烟包外观检测中去噪和边缘检测是两项关键的图像处理步骤,处理的结果与烟包外观质量特征提取、模型搜索等有很大关系。因此本文主要基于小波变换从烟包外观去噪和边缘检测两方面展开研究。本文首先阐述选择课题的原因以及研究的意义、以及目前烟包图像检测的状况与未来发展的趋势。然后重点讲解了小波理论,对比了几种传统去噪方法,以及软硬阈值法小波去噪。根据文献提出一种自适应小波阈值去噪算法,克服了常用小波阈值去噪算法中硬阈值函数存在间断点和软阈值函数信号输入输出有偏离的劣势。实验结果表明这种算法较好的存留了图像细节,去除干扰的效果较理想。其次,对几种常用的边缘检测算法进行了 matlab仿真,结果说明这几种方法或多或少存在一些不足,要么会丢失重要的边界信息,要么会受到噪声的影响。因此,本文运用二进小波变换的模极大与信号多尺度检测出烟包外观不同部分的边界,采用改善的Otsu法作为遗传算法的适应度函数,寻优找出分割门槛值,并运用到烟包外观检测,结果满足实际生产应用。最后对小波变换在烟包图像检测中的应用这个课题进行了总结,并在今后的烟包图像检测工作中强化学习方法理论,弥补不足,发现新课题,解决更多实际问题。更多还原

【Abstract】 The appearance of cigarette bag affects the quality of product directly,and quality is the life of enterprise.Quality is the key to maintain a certain market share in the increasingly fierce cigarette market.The appearance of the process quality of cigarette packet including trademark package is in place or not,damage,reverse package,scratch,plastic etc.;tax cut is in place or not,regular paste position or not,breakage,bend angle etc..Noise detection and edge detection are two key image processing steps in cigarette appearance inspection.The results of processing are closely related to appearance quality,feature extraction and model search.Therefore,this paper mainly studies the two aspects of appearance,denoise and edge detection of cigarette packets based on wavelet transform.This paper first describes the reasons for choosing the topic and the significance of the research,as well as the current status and future trends of cigarette package image detection.Then,the wavelet theory is emphasized,and several traditional denoise methods are compared.According to the literature,an adaptive wavelet threshold denoise algorithm is proposed,which overcomes the disadvantage of the existence of discontinuous point and soft threshold function in the usual wavelet threshold denoise algorithm,and the input and output signals deviate from each other.The experimental results show that the algorithm has better image details,and the effect of removing interference is ideal.Secondly,on several edge detection algorithm for the MATLAB simulation results show that these methods have many shortcomings,or lose the important boundary information,or be affected by noise.Therefore,this paper uses two wavelet transform modulus maxima and multiscale detection of cigarette appearance in different parts of the boundary,uses the improved Otsu method as the fitness function of genetic algorithm,optimization to find the segmentation threshold,and applied to cigarette appearance detection,the results meet the actual production application.Finally,the wavelet transform in cigarette package application of image detection in this paper was summarized.And in the future cigarette appearance detection work,strengthen learn methods and theory,make up for deficiencies,research new issues,solve more practical problems.更多还原