Facial Expression Recognition method based on residual network and attention mechanism

As a primary medium for emotional expression, human facial expressions carry rich informational value. Recent advancements in residual networks and attention mechanisms have broadened their application in expression classification, yet challenges persist in suboptimal key feature extraction and complex model training. To address these issues, this study proposes a novel facial expression recognition method integrating residual networks with attention mechanisms. The framework employs ResNet50 as the backbone network for feature extraction, enhanced by the Convolutional Block Attention Module (CBAM) to autonomously learn and prioritize critical features. Further innovations include reconstructing residual modules within the backbone network to optimize feature extraction and introducing a CAM-adjusted CBAM-ERF mechanism to mitigate neuronal suppression in specific regions, thereby accelerating network convergence and classification efficiency. Experimental results demonstrate the proposed residual network achieves 73.45% and 96.97% accuracy on the FER2013 and CK+ datasets, respectively.