基于图像处理技术的面粉麸星检测技术研究与实现

【作者】 乔常明；

【导师】 孙来军；

【作者基本信息】 黑龙江大学 ， 检测技术与自动化装置， 2010， 硕士

【摘要】 面粉中麸星含量的多少是评价面粉等级的一个重要指标,它不仅能够反映小麦品种的优劣,而且能够体现出面粉生产工艺的水平。目前国内绝大多数与面粉制品有关的科研院所、生产厂家等检测机构还是采用操作人员在放大镜的帮助下肉眼识别的方式来完成检测工作,这种传统的人工检测方式往往达不到精度要求而且费时费力,因此检测结果很难令人信服。针对目前存在的这些问题,本文提出设计一个智能的面粉麸星检测系统,该系统可以自动完成面粉装载、运输、拍照等一系列工作,同时该系统的上位机软件可以实时的完成对面粉图像的处理和麸星的快速检测,从而有效避免了传统人工检测方式中存在的诸多不足。在麸星检测系统的硬件设计中,将光照度检测及控制技术添加到检测仪器中,从而避免了其它仪器在检测时缺少对于检测环境实时监控的不足。在此基础上,为了满足光照度准确检测的设计要求,研究运用改进的BP神经网络来实现A/D采样数据与标准光照度值之间的数据拟合。仿真实验表明：这种改进的BP算法可以有效的避免局部极小化问题,提高了网络的收敛速度。本文还对图像灰度化、二值化、平滑滤波、纹理分析、边缘检测等多种数字图像处理技术进行了综合的比较研究,并参考模式识别的相关理论,提出了一种新的麸星快速识别和归一化的算法。然后,根据这些图像处理算法的基本原理设计编写了一款软件,该软件可以完成麸星个数、面积、面积百分比等参数的快速有效检测。最后,本文指出了该系统目前存在的一些不足,有待于日后改进和完善。更多还原

【Abstract】 The number of bran specks contained in the flour is an import indicator to evaluate the rank of flour, which not only reflects the pros and cons of wheat varieties, but also reflects the level of the flour production process. Currently, operators in most of the scientific research institutes, manufacturers and other testing organizations related with flour products, still use the way of visually identified with the help of a magnifying glass to complete the inspection work, which was the traditional manual testing method often do not meet accuracy requirement and time-consuming effort, so the test results is hardly convincing.In response to these problems, this paper presents to design and develop an intelligent measure system of bran specks in flour. This system can automatically complete flour loading, transportation, photograph and other works, while the Host Computer Software can complete real-time image processing and fast detection of bran, effectively overcomes the shortcomings in traditional manual testing method.In the hardware design of bran specks measure system, the technology of illuminance detect and control is added to the instrument, and thus avoiding the deficiency of lack of real-time monitoring for testing environment in other instruments. On this basis, in order to satisfy the design requirements of accuration for illuminace detection, research to apply improved BP neural network to achieve data fitting between A/D sampling data and the standard values of illuminace. The simulation result shows that the improved BP algorithm can effectively avoid the local minimization problem, and improve the network convergence speed.This paper also implements a comprehensive comparative study with a variety of digital image processing technology, such as, the image gray-based, binarization, smoothing filtering, texture analysis, edge detection and so on. After the reference to the relevant pattern recognition theory, a new bran specks rapid identification and normalization algorithm is presented. Then, designs an intelligent software based on these basic principles of image processing algorithms. This software can quickly detect number, area, area percentage and other parameters of bran specks. Finally, the paper also proposed some shortcomings in the system which to be improved and perfected in the future.更多还原