Terahertz spectrum identification method based on BLSTM-RNN

Abstract

The invention relates to a terahertz spectrum identification method based on BLSTM-RNN, and belongs to the technical field of spectrum analysis and substance type detection. The method comprises the following steps that firstly, a terahertz spectrum data set is filtered and denoised, then cubic spline interpolation is conducted on a frequency spectrum curve, and data are intercepted in the comparable same frequency range for resampling, so that data normalization processing is completed. According to the method, a BLSTM-RNN model is built to perform full-spectrum information automatic featureextraction on training set samples, a time reverse propagation algorithm and an Adam optimization algorithm are used for carrying out multiple iterative training on the model, and finally high-precision identification and classification of the test set are realized; and automatic feature extraction and effective classification can be realized on a high-dimensional terahertz spectrum data set, so that the problems that an existing terahertz spectrum identification method is tedious in process and high in technical requirement due to the fact a small number of key frequency spectrum features arerequired to be extracted firstly and then identification and classification are conducted for improving the classification precision are avoided.

Description

technical field [0001] The invention relates to a terahertz spectrum recognition method based on BLSTM-RNN (bidirectional long short term memory-recurrent neural network, bidirectional long short term memory recurrent neural network), which belongs to the technical field of spectrum analysis and substance type detection. Background technique [0002] In the field of material identification, spectroscopy plays a very important role. Among them, molecular vibration analysis techniques such as near-infrared spectroscopy and Raman spectroscopy have developed rapidly, and they use the characteristic spectra exhibited by substances for qualitative identification and quantitative analysis of substances. The terahertz (THz) spectrum in the far-infrared band also has "fingerprint" characteristics, and the terahertz band has perspective, security and spectral resolution capabilities. These characteristics make the application of THz technology in material identification and non-destr...

AI Technical Summary

Problems solved by technology

Although using PCA can effectively eliminate redundant information, it is necessary to manually select the principal components to be retained when obtaining terahertz spectral features, and the principal components with small contribution rates may contain important information on sample differences, which is not conducive to similar data sets. Subsequent identification

SVM is suitable for classification of small samples and low-dimensional data. Its disadvantages are that it is difficult to determine the parameters and the amount of data calculation is too large.

The calculation method of fuzzy recognition is relatively simple, but it often relies heavily on good feature engineering

Therefore, the above-mentioned terahertz spectrum identification method has problems such as cumbersome identification process and high technical requirements for feature extraction.

Images