A New Method of Mixed Gas Identification Based on a Convolutional Neural Network for Time Series Classification

Area of Science:

• Artificial Intelligence
• Chemical Sensing
• Machine Learning

Background:

• Convolutional Neural Networks (CNNs) excel in computer vision but are underexplored for gas data classification.
• Existing methods for mixed gas identification face limitations in speed and effectiveness.
• Time series data from gas sensors presents unique challenges for traditional classification models.

Purpose of the Study:

• To propose and evaluate a novel method for mixed gas identification using CNNs.
• To adapt CNNs for classifying time series data from an array of metal-oxide-semiconductor (MOX) gas sensors.
• To explore the efficacy of mapping time series data to image-like matrices for CNN analysis.

Main Methods:

• Time series data from eight MOX gas sensors detecting five mixed gases were collected.
• A novel approach mapped time series gas data into analogous image matrix data.
• Five established CNN architectures (VGG-16, VGG-19, ResNet18, ResNet34, ResNet50) were employed for classification.
• CNN parameters were fine-tuned to optimize gas recognition performance.

Main Results:

• The proposed method successfully classified five types of mixed gases using CNNs.
• The optimized CNN models achieved a final gas recognition rate of 96.67%.
• The approach demonstrated rapid and effective classification of gas sensor data.

Conclusions:

• The study successfully combined gas time series data with classical CNNs for mixed gas identification.
• Mapping time series data to image matrices is a viable strategy for applying CNNs to gas sensing.
• This method offers a promising new direction for advanced mixed gas identification systems.