<p>Traditional face recognition systems deployed on mobile and embedded devices often suffer from limited computing power and high energy consumption. To address these challenges, this study proposes a real-time face recognition algorithm based on a lightweight neural network, called the Light Face Network, aiming to improve detection speed, recognition accuracy, and energy efficiency under constrained hardware environments. The model integrates depthwise separable convolutional layers, channel-attention mechanisms, and a feature-compression strategy to reduce computational complexity while maintaining precision. Experimental results show that the proposed network achieves a recognition accuracy of 97.1% for anger emotion and maintains an overall accuracy above 90% on public benchmark datasets including Labeled Faces in the Wild and Age Database. Under partial occlusion scenarios such as sunglasses and scarf coverings, the recognition accuracy reaches 98.8% on smartphone platforms and 95.7% on surveillance cameras, while maintaining low energy consumption levels of approximately 3550 mW and 3800 mW, respectively. These findings demonstrate that the lightweight neural-network-based model can achieve high-precision and energy-efficient face recognition in real-time applications, providing a robust foundation for intelligent monitoring and human-computer interaction systems.</p>