The Application of Face Recognition Method Based on Deep Learning in Robot Vision System

With the continuous breakthroughs in deep learning technology, face recognition methods based on deep learning have become a prominent research focus in the field of computer vision. In particular, within service robots and embedded intelligent systems, face recognition plays a critical role in identity verification, interactive control, and behavioral understanding, and its performance directly affects the intelligence level of such systems. However, traditional face recognition models based on deep convolutional neural networks (CNNs) are often computationally intensive and contain a large number of parameters, making them unsuitable for deployment on robotic platforms that require real-time processing, low power consumption, and lightweight models. To address these challenges, this paper proposes a lightweight face recognition method based on deep learning, which combines the MobileNetV3 architecture with the Convolutional Block Attention Module (CBAM) to construct an efficient recognition model suitable for robotic vision systems. MobileNetV3, as the backbone network, provides excellent computational efficiency and structural compression capabilities, effectively reducing the size and latency of the model, while the CBAM module introduces channel and spatial attention mechanisms to guide the network to focus on key facial regions during deep feature extraction, thereby enhancing the discriminative power and robustness of recognition. Extensive experiments are conducted on the publicly available Labeled Faces in the Wild (LFW) dataset, where the model is trained using the cross-entropy loss function and optimized with the Adam optimizer, evaluating its performance under realistic scenarios such as complex backgrounds, pose variations, and occlusions. Experimental results show that the proposed model achieves higher recognition accuracy than several existing lightweight networks while maintaining a compact structure, demonstrating better adaptability and generalization. This method effectively balances accuracy and real-time performance, offering a feasible and efficient solution for robot-oriented face recognition systems. The study confirms the effectiveness of integrating deep learning and attention mechanisms into lightweight architectures and provides new ideas and practical paths for achieving high-performance face recognition on edge computing devices, with strong application potential.