Design of face recognition system based on deep learning

With the popularity of biometric technology, this paper designs a face recognition and execution control system based on Raspberry Pi and STM32. The system realises image acquisition and deep learning recognition through Raspberry Pi, and combines with STM32 hardware control module to form closed-loop operation. Aiming at the limitations of traditional methods, the innovative fusion of LBP features and deep neural networks, and the use of support vector machines to complete the classification task. Experiments show that the system has an accuracy of 99.05% on the LFW dataset, which is better than mainstream algorithms such as DeepFace and FaceNet. It is verified that the system maintains the lowest recognition rate of 91.3% under normal light, backlight and dark conditions, with remarkable environmental adaptability. In small sample training scenarios, this method improves the accuracy by 3% compared with the comparison algorithms and shows stronger generalisation ability. The study shows that the system combines high accuracy, strong robustness and hardware scalability, and is suitable for security, intelligent interaction and other real-world scenarios. In the future, we will focus on optimising the computational efficiency of the algorithm and enhancing the anti-interference ability in complex dynamic environments.