Enhanced DeepID Network-Based Access Control for Property Management Using Transformer and Gaussian Mixture Models
Huiqi Zhang · Informatica · 2025
In recent years, property backend access control systems have faced many challenges in terms of management efficiency and security. Traditional authentication methods are difficult to cope with complex and changing access scenarios, leading to security vulnerabilities. To address this issue, a property backend access control model based on an improved DeepID network is designed. The Transformer model is introduced to optimize the feature extraction capability of the DeepID network, and the multi-head attention mechanism is applied to optimize feature expression and improve the accuracy of face feature recognition for users. Moreover, a Gaussian mixture model is introduced to accurately model user behavior patterns. The experimental study was conducted on a platform with Intel Core i9-12900K, NVIDIA RTX 3090, and 64GB RAM, using 13,233 face images from LFW dataset for training and 5,000 images from CelebA dataset for validation. According to the results, the improved DeepID network model achieved a feature recognition accuracy of 97.3%, with a loss value of only 0.15, statistically significantly outperforming traditional DeepID network (86.7%), VGG-Face (92.4%), and FaceNet (94.9%) in terms of F1-score and precision-recall metrics. The research provides an efficient and reliable technical solution for the property backend access control system, which has important practical significance for improving the intelligence and security of property management.