Facial emotion recognition is still a challenging task in computer vision because human facial expressions are very subtle and complex. In this paper, we address this issue and propose a novel deep-learning framework that combines multi-scale feature extraction with a dynamic attention mechanism and improved residual connection. The research aims to create a reliable system that identifies facial expressions correctly in different circumstances. The proposed method was validated rigorously on a standard face expression recognition data set, with an impressive overall accuracy of 96.1%. Additionally, the model performed remarkably well on extra metrics like precision, recall, and F1-score. These findings highlight the model’s ability to learn and distinguish subtle features in human faces, leading to improved performance compared to conventional methods. In summary, this research makes a noteworthy contribution to affective computing by paving the way for the future development of real-time systems that can recognize human emotions, enabling numerous potential applications in the fields of mental health assessment, human-computer interaction, and adaptive user interfaces.