Меню
Vol. 2 No. 31 (2026) / 2026-06-30 / Number of views: 95
Authors
Keywords
Link to DOI:
How to quote
This article addresses the problem of automatically identifying the composition of gas mixtures based on data from a multispectral sensor. The analysis of gaseous media is a critical task for industrial safety, environmental monitoring, medical diagnostics, and intelligent sensor systems. Traditional gas analysis methods based on spectroscopy provide high measurement accuracy but require complex equipment and labor-intensive data processing. In this regard, the application of artificial intelligence methods for processing spectral signals is a timely area of research. The aim of this work is to develop and investigate a convolutional neural network model for identifying gas mixtures based on spectral data from the AS7265x multispectral sensor. A laboratory setup was created for conducting experiments, including a multispectral sensor and a system for controlling experimental conditions based on temperature and pressure sensors. Spectral vectors containing 18 intensity channels corresponding to different wavelengths were used as input data. To process the spectral data, a one-dimensional convolutional neural network (1D-CNN) architecture was developed, capable of identifying characteristic spectral features of various gases. The model was trained and tested on an experimental dataset containing measurements of air, carbon dioxide, and a mixture of oxygen and nitrogen. The experimental results showed high classification accuracy, reaching 100% on the test set. The results confirm the effectiveness of deep learning methods for analyzing spectral data and demonstrate the potential for using compact multispectral sensors in intelligent gas analysis systems.