Fine-Grained Micro-Expression Recognition Based on Hierarchical Attention Mechanism

Micro-Expressions can significantly reflect the psychological state of people at the current moment and have great research value. In the context of the continuous development of deep learning technology, the use of deep learning algorithms to solve the problem of micro-expression recognition(MER) has achieved obvious advantages, but there are also many problems. To address the problem of fine-grained feature extraction for micro-expressions, we propose a hierarchical attention mechanism-based micro-expression recognition(MER) model, which consists of a single ResNet18 network and the task-specific attention modules. We use the ResNet18 network as the backbone network, and insert the attention modules among multiple intermediate layers of the network for extracting fine-grained micro-expression features, so that the model focuses on task-related regions and ignores task-irrelevant regions. We designed a micro-expression recognition(MER) experiment to verify the model. Under the dual evaluation criteria of UF1 and UAR, our proposed model can reach 81.99% and 77.93%, respectively, which has achieved significant performance improvement compared to the current mainstream models.