Leveraging Spatial Information with MobileViT for Enhanced Microwave Measurement

Accurate measurement of pulverized coal concentration (PCC) is a critical and complex task in coal-fired systems. Traditional microwave attenuation methods often emphasize numerical analysis while neglecting the spatial information, making them susceptible to the nonuniformity of pulverized coal flow. To address this issue, we propose a novel MobileViT-based method that leverages the spatial features of microwave sensor. Instead of using directly the attenuation signals, the method converts them into a two-dimensional bitmap and takes it as input for a lightweight MobileViT network. A cyclic shifting rule is then proposed for data augmentation, to help the network learning more consistent features from bitmaps. Numerous experiments involving 2, 8 and 16 electrode sensors validate the effectiveness of the proposed method in measuring the PCC. Furthermore, the proposed method significantly outperforms the method utilizing only one-dimensional data, with the recall rate improving dramatically from $\mathbf{6 0 \%}$ to $\mathbf{9 0 \%}$ under the 16 -electrode configuration.