14 / 2022-09-09 09:22:12

sEFER: sEMG-based Facial Expression Recognition Exploiting Bi-LSTM and Attention Mechanism

Facial expression recognition (FER) based on surface electromyography (sEMG) is a promising task for human emotion understanding. It is challenging due to the nonstationary nature of sEMG in real-life operation. To solve the problem, this paper proposes a sEMG-based FER method named sEFER, which can improve recognition accuracy by integrating bi-directional long short-term memory (Bi-LSTM) network and multiple head attention mechanism. To achieve better emotion recognition performance, the architecture of our method has two distinguish characteristics: First, multi-head mechanism solves the multi-channel data collection of our data and make the classification more efficiently. Second, the attention mechanism helps us focus more purposefully on certain channels during the data processing to improve recognition accuracy. We evaluate the proposed method on the data set which is constructed by obtaining high-fidelity sEMG signals through novel epidermal electrodes. Our method can achieve high average accuracy for emotion recognition. Qualitative and quantitative experiments illustrate that our method is superior to the performance of the classical facial expression classification methods and has strong robustness.