基于小波神经网络的光谱数据压缩与分类研究  

STUDY OF SPECTRAL DATA REPRESENTATION AND CLASSIFICATION FOR WAVELET NEURAL NETWORK

作　　者：张永胜[1,2] 郁可[1,2] 

机构地区：[1]武汉工业大学计算机科学系 [2]武汉工业大学光纤中心

出　　处：《计算机研究与发展》1999年第8期973-977,共5页Journal of Computer Research and Development

基　　金：国家自然科学基金

摘　　要：文中介绍了一种基于小波分析而构造的神经网络模型——小波神经网络．利用它并适当选取网络结构和小波基，实现了对化学物质红外光谱数据的压缩表达和分类．实验表明，网络在大幅度压缩数据的同时能很好地恢复原始光谱、较准确地反映吸收峰的位置和强度．在分类方面它比其它网络具有更高的分辨率和特征提取能力．通过与ＢＰ网络的训练结果对比，小波神经网络具有自适应性强、收敛速度快及可屏蔽随机噪声等优点，特别适用于非平稳、非线性信号的分析研究．因此，小波神经网络在光谱处理方面有着较好的应用前景和优越性．A model of neural network based on wavelet analysis wavelet neural network, is introduced in the paper here. The chemical substance infrared spectral compression representation and classification are realized with adaptive network structure and daughter wavelet.The experimental results show that the original spectra can be recovered well, the place and intensity of absorptive peaks can be defined exactly, and the data can be compressed greatly with the wavelet network.The resolution ratio and the characteristic collection ability of the wavelet network are better than that of other networks in classification. A comparison of the two training results of the wavelet and BP networks indicates that the wavelet network has better adaptability and faster convergence speed and can shield random noise. The wavelet network can be applied to the study and analysis of nonstationary and nonlinear signals, so it holds a bright future in spectra processing.

关 键 词：小波神经网络 数据压缩 分类 红外光谱 光谱 

分 类 号：O433.5[机械工程—光学工程]