Research on Facial Expression Recognition Based on YOLOv9

Facial expression recognition is an important means to achieve emotional recognition. To improve the emotional feature analysis ability of facial expression recognition, a YOLOv9 based facial expression recognition method is proposed. Using feature extraction and emotion classification methods to extract basic facial expression features, constructing a residual multi-scale feature fusion attention mechanism model to filter out high-quality features of subtle facial expression features, and sparsely reconstructing the selected features. Using the Smooth maximum unit activation function, the face is finely divided into several motion units. A deep learning model is used to extract facial texture features. YOLOv9 is used to dynamically freeze high-dimensional facial feature extraction networks. Gabor filtering is applied to the input image in four different directions. Based on the filtering results, cross entropy loss detection and expression classification of facial expressions are achieved. The simulation results show that using this method for facial expression recognition can effectively recognize features with small changes between expressions, and has strong sentiment analysis ability. It achieves a recognition accuracy of 96.877% on strong interference experimental datasets and has high robustness.