【作者】 薛锋；

【导师】 翁桂荣；

【作者基本信息】 苏州大学 ， 通信与信息系统， 2006， 硕士

【摘要】 在图像处理和模式识别领域经常根据物体的边界来识别物体或对其分类。实际应用中,对物体边界特征描述的方法很多,本文对数学形态学、相对矩、基于极半径函数的傅里叶子这几种方法进行了比较研究。发电机组的转轴的轴心轨迹形状是重要的故障征兆,在机组故障诊断中起着重要作用。在实际操作中,根据轴心轨迹图还可以得出故障的前期征兆,对防止故障的恶化和排除故障具有指导作用。本文主要介绍了以归一化数学形态学谱、相对矩和形状参数为特征,结合人工神经网络实现对轴心轨迹的自动识别和分类。选择改进型BP网络,经过学习和优化,建立轴心轨迹参数和特征谱之间的非线性映射。通过网络的非线性映射和联想,获取轴心轨迹的参数,以便分析出此时发电机组运转时的状况。实验得到的综合值和实际的理论值是很接近的,这表明本文所选取的特征谱有较好的区分能力;在网络训练成功之后,系统基本可以实现实时提取轴心轨迹的参数;网络还可以联想到学习样本库中以外的数据;由此可见,将该方法加入到故障诊断专家系统中,可实现诊断过程的自动化,大大方便故障诊断系统的使用,提高故障诊断系统的自动诊断水平及准确率。更多还原

【Abstract】 Description or discrimination of boundary curves is an important problem in image processing and pattern recognition.There are too many boundary feature description methods for object recognition,three main methods are Fourier descriptors,Relative moments, mathematical morphology.The performance of each of the methods was compared.Chart of axes track has the important malfunction information,it is very important in diagnostics of malfunction.In practice,we can obtain earlier stage malfunction sign of the generator that comes from the chart of axes track,and it would have instructive function to preventing deterioration and excluding malfunction.In this paper,we combine Normalized pattern spectrum, Relative mome-nts and Geometrical characteristic with artificial neural network to achieve classify and recognize the chart of axes track. Afterwards, neural network linked the non-linear relationship between the track’s parameters and it’s characteristics.The experiments data is very close to the ideal parameters,it also validate the effectiveness of the method. The neural network can get the track’s更多还原