CC BY
1On the possibility of using neural network in tasks of laser induced breakdown spectroscopy
A.V. Bulanov1
1- V.I. Il'ichev Pacific Oceanological Institute, FEB RAS, Vladivostok, Russia
a_bulanov@me.com
In the context of increasing human impact on ecosystems and the introduction of international carbon taxes, it becomes extremely important to study the flows, emissions and "burial" of carbon dioxide in various environments, as well as the creation of corresponding "carbon polygons". The use of optical spectroscopy methods allows for continuous monitoring of many environmental characteristics in real time, both directly on site and at a distance. To solve a number of fundamental and applied problems, regular measurements taken in the water column are necessary. The use of spark and laser-induced breakdown spectroscopy (LIBs) methods for elemental analysis of liquids in oceanological research is a relevant research, but is accompanied by certain difficulties.
To solve the problems of studying the World Ocean, an automated complex for studying the spectral optical characteristics and hydrophysical characteristics of the upper layer of the sea using the flow method was developed, which makes it possible to study the variability of the optical and hydrophysical structure of the marine environment along the vessel route crossing various water masses.
In order to record optical data, a specially developed spark complex was used. To analyze the obtained spectral data, a program was created in the Python programming language, designed for processing and visualizing statistical data of laser breakdown. The input data for the program was a set of *.csv data breakdown spectrum image files, generated using STM32 and transferred to a microcomputer via a serial port. The program indicated the wavelength of the monochromator when recording optical breakdown. Processing of files into folders with different values of exposure, delay, etc. was implemented. Preset analytical algorithms for optical image processing (averaging, finding peaks, etc.) were implemented. The result was a breakdown spectrum with highlighted spectral lines of chemical elements. As a result, the dependence of the intensity of the spectral lines of chemical elements, such as sodium and calcium, was obtained during optical breakdown in an aerosol cloud from sea water using ultrasound.
In addition to the above listed experimental methods for studying matter in order to improve the sensitivity of LIS spectroscopy, the problem of processing a large amount of data received in real time was solved. It should be noted that the rapid processing of numerous spectra obtained as a result of optical breakdown measurements is associated with detailed analysis and in a significant number of cases is a labor-intensive procedure. The rapid development of current trends related to the use of neural networks and machine learning algorithms could come to the rescue and allow more accurate classification, regression, clustering and other operations with samples to obtain information about the spectrum and study of matter. To further improve the sensitivity of the LIBs method, analysis of breakdown signals using artificial neural networks (ANN) was proposed, which was used to estimate the contribution of dissolved organic and inorganic carbon in a carbon test site.
The model was implemented in Python using TensorFlowTM, and the weights and biases of the model were optimized using a backpropagation algorithm.
To summarize, the obtained studies of laser breakdown made it possible to formulate the basic principles of creating a method of combined ultrasonic laser spark spectroscopy and to create a compact automated complex for studying the spectral optical and hydrophysical characteristics of the upper layer of the sea using the flow method. This complex was successfully tested under expeditionary conditions during voyage No. 81 of the R/V Professor Gagarinsky in the Sea of Japan in August 2022, as well as during the 52nd voyage of the R/V Akademik Boris Petrov in the Atlantic Ocean and in the plume of the Amazon River in October - December 2022. Using this complex, new data on the state of sea water with high spatial resolution were obtained.
