Dynamic Facial Expression Recognition (DFER) is considered a more challenging task in computer vision due to its closer to the emotional demands of the real world. The Spiking Neural P (SNP) system is a biomimetic model that imitates the functioning of the human brain and aligns with human perception, providing better interpretability for biological features. Therefore, based on these biologic characteristics, this paper has developed a Nonlinear Spike Neural P Network (NSNPnet) to extract sequential features, which is beneficial for the dynamic analysis of emotions. Additionally, introducing a Customized Enhancement Spatial Module (CESM) enhances NSNPnet’s capability to extract spatiotemporal information. The Spatiotemporal Convolution Transformer (STC-Former) identifies key features in each video frame and models inter-frame context, extracting spatiotemporal representations. Combining these modules allows the Nonlinear Spiking Neural P Network For Dynamic Facial Expression Recognition (NSNP-DFER) to identify facial expressions in video sequences accurately. The model is experimented with and evaluated on two benchmark databases: FERV39K and DFEW. The results of comprehensive experiments on these two databases show that the proposed method can achieve robust facial expression recognition outcomes in real-world scenarios.

中文翻译：

---

NSNP-DFER：用于动态面部表情识别的非线性尖峰神经 P 网络

动态面部表情识别（DFER）被认为是计算机视觉中更具挑战性的任务，因为它更接近现实世界的情感需求。Spiking Neural P (SNP) 系统是一种仿生模型，模仿人脑的功能并与人类的感知保持一致，为生物特征提供更好的解释性。因此，基于这些生物学特征，本文开发了非线性尖峰神经P网络（NSNPnet）来提取序列特征，有利于情绪的动态分析。此外，引入定制增强空间模块（CESM）增强了 NSNPnet 提取时空信息的能力。时空卷积变换器（STC-Former）识别每个视频帧中的关键特征并对帧间上下文进行建模，提取时空表示。结合这些模块，非线性尖峰神经 P 网络动态面部表情识别 (NSNP-DFER) 可以准确识别视频序列中的面部表情。该模型在两个基准数据库：FERV39K 和 DFEW 上进行了实验和评估。对这两个数据库的综合实验结果表明，所提出的方法可以在现实场景中实现鲁棒的面部表情识别结果。