摘要
将小波变换和神经网络相结合用于非线性荧光光谱的识别,针对非线性荧光光谱的特点,提出了选择最佳小波函数和分解层数的方法,处理后的光谱在保留光谱特征的基础上,大大压缩了数据维数;采用概率神经网络(PNN),对3种污染气体的非线性荧光光谱进行识别,获得了满意的实验结果。由于神经网络的输入是小波压缩后的数据,不仅提取了原始数据中的特征,而且数据的维数也下降7倍多,大大提高了气体识别的速度。
A combination of optimizing wavelet transform and neural network is applied to recognizing nonlinear fluorescence spectra. The optimization wavelet function and decomposition layers are proposed. A probabilistic neural network (PNN) model is employed in order to recognize the nonlinear fluorescence spectrum of 3 impurities in the air, and satisfied experiment results have been acquired. The operation rate of the network has been greatly enhanced because the optimal input data obtained after the wavelet transform are not only the features of original signals, but also consumedly compressed, so that the dimension of the data becomes much less than that of original signals.
出处
《光电子.激光》
EI
CAS
CSCD
北大核心
2005年第6期718-721,共4页
Journal of Optoelectronics·Laser
基金
国家自然科学基金资助项目(60277022
60477009)
天津市自然科学基金重点资助项目(023800811)
博士点基金资助项目(20030055022)
南开大学科技创新基金资助项目