Research on Driver Facial Expression Detection Algorithm Based on BBAN-Net

With a variety of uses in human-computer interaction, safe driving, and other domains, driver expression recognition is used to identify the type of expression from face photos. Various labels for expression data are frequently unsegregated in real-world applications. Expression recognition model performance is impacted when expression labels are not aggregated, as this lowers classification optimizer accuracy. Because of this, this study suggests a facial expression recognition technique called BBAN-Net for expression label non-aggregation. In the feature learning stage and the classifier optimization stage, respectively, this technique optimizes the model for category aggregation. BBAN-Net achieves distributed aggregation of labels for each category by combining the BBN and ANFL structures with a dual-branching architecture that uses regular learning and rebalancing. Before and after the addition of the expression label pre-training module BBN was verified on DISFA data, the experimental results showed that for expression label pre-processing, the F1-Score improved by 31.16%. Quantitative experiments conducted in four different networks show that BBAN-Net achieves 67.43% accuracy in detecting the average F1- Score on 8 AUs with 81.7% accuracy and 85.9% AUC. It also compares relatively well with a host of other state-of-the-art methods.