A New Try for Age Estimation and Gender Recognition Based on EfficientNetB4

A framework called ECA-EfficientNetB4 is proposed to solve the issue of low accuracy in age estimation and gender recognition of current facial features in this paper. First, integrates ECA attention mechanism in EfficientNetB4, which can effectively capture inter channel relationships in images while maintaining efficiency, thereby enhancing the ability of feature representation. After adding the cross entropy loss function to the network, not only does it enhance the accuracy of prediction, but also enhances the convergence speed of the model. The ECA-EfficientNetB4 model proposed in this article has an age MAE of 4.55 years on a UTKFace dataset of over 23000 images, which is 0.31 years less than VGGFace-ResNet50 and the accuracy rate of gender recognition is 93.62%. Therefore, compared with the other four excellent frameworks in the same category, this model has the lowest MAE for age and achieves first-class accuracy for gender, thus achieving excellent performance. In the future, this technology can be applied to identity verification and assisted driving judgment strategies on cars.