Research on the Performance of Multiple Deep Learning Models in Facial Age Recognition

Facial age recognition, as an important research in the field of computer vision, has broad application prospects in areas such as security monitoring and human-computer interaction. This study aims to compare the performance of classic (VGGNet, GoogleNet, DenseNet) and lightweight (MobileNet, EfficientNet) deep learning frameworks in facial age recognition, providing a scientific basis for model selection. The research is based on the public datasets MORPH (large-scale) and FG-NET (small-scale), comprehensively evaluating from dimensions of accuracy, model efficiency (parameter scale, training time), and resource consumption. Experimental results show that there is no significant linear relationship between model parameters, accuracy, and training time. DenseNet121 achieves the best accuracy with MAE is 2.50 on MORPH, suitable for high-precision requirements; GoogleNet performs best on FG-NET with MAE is 4.02; MobileNet and other lightweight models are suitable for mobile devices but have slightly lower accuracy than large models. This study quantifies the adaptation characteristics of different models, providing core evidence for the selection of scenario-based face age recognition models, facilitating the practical application of this technology.