ECA-Based Dynamic Data Enhancement Model for Facial Expression Recognition

Facial expression recognition is an important task in affective computing and human-computer interaction. This paper proposes a lightweight deep network for real-time expression recognition that balances accuracy and computational efficiency to address the challenges of recognition in complex environments. Therefore, this paper proposes a dynamic data augmentation model based on ECA. This model integrates the Efficient Channel Attention (ECA) module into the ResNet-18 architecture. We need to optimize channel interaction through adaptive 1D convolution and combine dynamic data augmentation strategies (random rotation, color jitter, Gaussian blur) to simulate real-world interference. Specifically, the ECA module enhances key feature responses through local cross-channel interactions, and Gaussian blurring simulates low-resolution scenarios to improve model robustness. This study was conducted on the Kaggle website FER dataset. The experimental results show that the proposed model improves the accuracy by 34.06% over the traditional CNN model. The model performs well in recognizing expressions such as “disgust” and “fear”, but has limited ability to generalize to micro-expressions and light-sensitive categories. This study provides efficient solutions for real-world scenarios such as mental health monitoring and intelligent robots, and reveals the optimization direction of the lightweight attention mechanism.