Sample Classification and Recognition Method Based on Convolutional Neural Network

Abstract

The embodiment of the present invention provides a sample classification and identification method based on convolutional neural network. During the training process of convolutional neural network, on the basis of the original training data, a certain amount of training data is obtained by arranging and combining, and then based on In practical applications, the characteristics of the difference between the spectral data collected by different equipment and different environments for the target substance and the actual spectral data are different. By adjusting the wavelength and intensity of the training data, it is possible to simulate the spectral data collected by different equipment and different environments. Spectral data, so that the trained convolutional neural network is suitable for the classification and identification of spectral data collected by different equipment and in different environments, so that there is no need to use standard substances for calibration; in addition, through random combination of original training data and simulated adjustment The method increases the amount of data, which can reduce the amount of original training data collection, that is, to achieve the expected convolutional neural network recognition effect in the case of small samples.

Description

technical field [0001] The invention relates to the technical field of laser-induced breakdown spectroscopy, in particular to a method for classifying and identifying samples based on a convolutional neural network. Background technique [0002] Laser-induced breakdown spectroscopy (LIBS, laser-induced breakdown spectroscopy) technology is a technology that uses ultra-short pulse laser to focus on the surface of the sample to form plasma, and then analyze the spectral data emitted by the plasma to identify the type of material it belongs to. [0003] However, in the actual application of laser-induced breakdown spectroscopy to identify substance categories, due to the influence of the laser-induced breakdown spectroscopy equipment itself and the environment in which the equipment is located, there is a gap between the detected spectral data and the actual spectral data of the substance to be detected. There are differences, and the difference is mainly reflected in the devia...

AI Technical Summary

Problems solved by technology

[0003] However, in the actual application of laser-induced breakdown spectroscopy to identify substance categories, due to the influence of the laser-induced breakdown spectroscopy equipment itself and the environment in which the equipment is located, there is a gap between the detected spectral data and the actual spectral data of the substance to be detected. There are differences, and the difference is mainly reflected in the deviation between the wavelength indicated by the abscissa and the intensity indicated by the ordinate between the detected spectral data and the actual spectral data. In this case, if the detected spectral data is used to identify If there is a discrepancy between the detected spectral data and the actual spectral data of iron ore, iron ore may be classified as non-ferrous ore

[0004] In related technologies, in order to enhance the recognition accuracy, standard substances are usually used for calibration before identifying the substance category. For example, the traditional peak position calibration method is to detect standard substances according to the characteristic peaks in the actual spectral data of the standard substances. wavelength to calibrate the wavelength of the detected spectral data, but this method can only ensure the accuracy of the wavelength at the characteristic peak used for calibration, while there are still different degrees of wavelength errors in other wavelength ranges, which still have a great impact on the recognition accuracy. Negative Effects

Images