Spiking Broad Learning System for Facial Expression Recognition in the Wild

Facial expression recognition (FER) has become an active field of research in recent years. Deep learning (DL) technologies have achieved excellent results in this field, but it still encounters numerous challenges due to its inadequate robustness and real‐time capabilities. The broad learning system (BLS) is a flat network structure that can be strengthened by expanding more feature nodes and enhancement nodes in wide sense, which proves to be beneficial for FER. However, the shallow network may limit the effectiveness of the feature learning performance. For this issue, this paper proposes a novel method that introduces neuromorphic spikes into BLS to form spiking BLS, which improves the feature representation capability. To demonstrate the efficiency of the proposed architecture network, we evaluate the spiking BLS on three benchmark datasets for wild facial expression classification: FERPlus, RAF‐DB, and SFEW 2.0. The experimental results show that the spiking BLS can achieve higher classification accuracy with the performance of 85.87% (FERPlus), 86.83% (RAF‐DB), and 55.85% (SFEW 2.0).