Spectral Channel Attention for Skeleton Action Recognition in Lubricant Filling

In Lubricant Filling, detecting the standard sequence of actions and the completion of key tasks is crucial to prevent issues such as missed filling or improperly tightened bolts. This paper proposes a computer vision-based solution, the Multi-Spectral Channel Attention Hybrid Frequency-aware Mixed Transformer (MSCA-FreqMixFormer), which adaptively captures key frequency channel responses during the feature fusion phase, optimizing the expression and coupling of spatial-temporal frequency features. Experiments and comparisons were conducted on a self-built dataset of standard operational actions, which show that the MSCA-FreqMixFormer outperforms other models with an accuracy rate of 98.96%.