【作者】 彭波；

【导师】 李旭宇；

【作者基本信息】 长沙理工大学 ， 机械电子工程， 2010， 硕士

【摘要】 颜色的测量和控制,尤其是色差的评定是科学研究和生产中经常遇到的问题。为了不断地适应实际工业生产的需求,要求能够对色样颜色进行准确测量并对其色差进行科学的评价,使其评价结果能够更贴切地反映色样间的颜色视觉差异,从而正确地判别产品的颜色质量,并能够由此达到对生产过程科学有效的控制。论文采用OPB780颜色传感器结合Elman神经网络和聚类算法对颜色识别进行了系统研究,取得如下进展:(1)介绍了物体和颜色的相互关系及相关颜色理论,分析了颜色测量的原理,讨论了颜色的三刺激值和色度值的转换、RGB颜色空间至CMYK颜色空间的互换。(2)分析了OPB780颜色传感器的工作原理和特性,设计了以单片机AT89C51为核心测量单元的外围辅助测量电路;考虑到对测量后数据显示的快速和稳定,设计了以PL2303芯片为核心的虚拟串口电路计算机显示方案。(3)根据OPB780颜色传感器的感光特性设计了光源补偿电路和数据采集暗室;根据白平衡原理和单片机频率测量原理,编制了颜色测量KeilC51软件程序。根据设计好的电路和软硬件测量出了2200个训练色卡样本和39个测试色卡样本的R、G、B、C频率值。(4)提出了基于Elman神经网络用于颜色识别的研究方法,通过对输入数据的归一化处理,并经过多次网络节点的测试调整,选取颜色识别效果最好的网络结构。在最终的色差聚类算法上,采用改进的K-means聚类算法:以黑蓝绿青红紫黄白八种常见颜色的RGB值为聚类中心点,聚类准则利用CIE1976Lab均匀颜色空间及色差公式,并按照色差最小原则确定出同这8个中心的聚类。通过2200个训练颜色样本数据和39个测试颜色样本数据用于色差聚类的试验结果表明,所构建的Elman神经网络具有良好效果,达到预期目的。更多还原

【Abstract】 The measurement and control of color especially the assessment of aberration are problems often encountered in scientific research and production. In order to adapt to the needs of the actual industrial production, it is required that the color of the sample colors should be accurately measured and the aberration should be evaluated in such a scientific way that the evaluation result can reflect more appropriately the visual differences between samples, helping to judge correctly the color quality of products and control the productive process scientifically and efficiently.This paper make a systematic study of color identification in a way that combines OPB780 color sensor with Elman neural network and clustering algorithm. The details are as follows:(1)This paper describes the relationship between objects and colors, theory of relative colors, as well as the principles for color measurement, it also discusses the conversion between the color tristimulus values and chroma values and the exchange from RGB color space to CMYK color space.(2) Through the analysis of OPB780’s color sensing principle and features, we designed external auxiliary measurement circuit with AT89C51 microcontroller as the core measurement unit; After taking into account the measured data show rapid and stable, designed to PL2303 chip circuits as the core of a virtual serial port computer display program.(3) We designed light compensation circuit and data collection chamber according to the OPB780 sensor photosensitivity color and programmed the Keil C51, software for color measurement, based on the white balance theory and the principle of frequency measurement of microcontroller. We have measured the R, G, B, C frequency of the 2200 training samples and 39 test samples of color card by the circuit and the software.(4) We proposed the methods for color recognition based on the memory function of Elman neural network. After the normalization of input data, and several adjustment of test network node, we designed the best network structure. Take the improved K-means clustering algorithm: calculating the color difference between target color and 8 common colors, i.e. black, blue, red, violet, blue, green, yellow and white, by use of CIE 1976L *a * b * color difference formula we proposed the evaluation criteria of clustering algorithms, that is , one color belongs to one of the common colors with which it has a minimum difference.The research results speak for the effects of Elman neural network.更多还原