Facial Gender Detection Based on Improved ResNet-18 in Multi-Resolution Scenario

Automatic gender detection is crucial in behaviour analysis and security surveillance applications, demanding effective algorithms for accurate target identification. This research proposes a gender recognition model based on a Convolutional Neural Network (CNN) architecture, specifically leveraging ResNet- 18. The model incorporates residual techniques to enhance feature map quality, thereby improving prediction performance. It utilises the UTKFace dataset, containing face images annotated with gender labels, was used to train the model. Experimental results demonstrate that the developed model achieves a validation accuracy of 91% and a validation loss of 0.7232, indicating a strong capability in gender identification from facial images. While these results are promising, there remains potential for further refinement. This research contributes to the development of a more efficient and accurate gender detection system through the implementation of a lightweight and practical CNN architecture. The model's feasibility for real-time applications suggests it can effectively distinguish gender in various operational scenarios.