论文部分内容阅读
为了解决红外光谱定量分析中的特征提取和校正规模问题,提出了一种输入层自构造神经网络。这种网络能够利用训练数据的某些先验知识,自动选择输入层神经元的个数。在学习过程中,输入神经元个数从最小值1开始,根据网络误差的变化逐步增加,最终确定最佳神经元数量。这种网络模型将特征提取和参数学习过程融为一体,有利于提高建模效率。利用仿真红外光谱的定量分析实验表明,这种网络模型不仅能够对光谱数据实现高效率的波长选择,并具有抑制随机噪声和非线性干扰的能力。
In order to solve the problem of feature extraction and calibration in quantitative analysis of infrared spectroscopy, an input layer self-organizing neural network is proposed. This network can use some prior knowledge of training data to automatically select the number of neurons in the input layer. During the learning process, the number of input neurons starts from the minimum value of 1 and gradually increases according to the change of network error, so as to finally determine the optimal number of neurons. This network model combines the feature extraction and parameter learning process, which helps to improve the modeling efficiency. Quantitative analysis using simulated infrared spectroscopy shows that this network model can not only achieve high efficiency wavelength selection of spectral data, but also reduce the random noise and nonlinear interference.