Scholarlink

Background <p>The increasing demand for secure and reliable identity verification has accelerated the adoption of biometric recognition systems that integrate multiple biometric traits to overcome the limitations of traditional biometric authentication methods. Conventional biometric systems often rely on static fusion techniques, which may fail to adapt effectively to varying environmental conditions, noise, and modality reliability. Therefore, intelligent and adaptive fusion strategies are essential for improving recognition accuracy and robustness in security-critical applications.</p> Objective <p>This research aims to develop a Deep Reinforcement Learning (DRL)-based identity recognition framework that adaptively fuses facial, iris, and fingerprint biometric modalities to enhance authentication performance under challenging real-world conditions.</p> Methods <p>The proposed framework collects biometric data from publicly available datasets and controlled experimental settings. Preprocessing techniques, including image normalization, alignment, and noise reduction, are applied to improve data quality. Convolutional Neural Networks (CNNs) are employed to extract discriminative feature representations from facial, iris, and fingerprint images. A Modified Lévy Flight Distribution-enhanced Deep Deterministic Policy Gradient (MLFD-DDPG) algorithm is introduced to perform adaptive fusion by dynamically optimizing modality weights through reward-driven learning. The fused feature vectors are subsequently classified using a fully connected neural network with a softmax classifier.</p> Results <p>Experimental evaluations conducted in Python on benchmark biometric datasets demonstrate that the proposed MLFD-DDPG framework achieves a recognition accuracy of 98.73%, outperforming conventional static fusion approaches in robustness, adaptability, and recognition performance.</p> Conclusion <p>The proposed DRL-driven biometric system provides an intelligent, scalable, and highly reliable identity recognition solution suitable for advanced security and authentication applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multimodal identity recognition technology based on Deep Reinforcement Learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月21日

查看全文
  • Huihong Xu

摘要

Background

The increasing demand for secure and reliable identity verification has accelerated the adoption of biometric recognition systems that integrate multiple biometric traits to overcome the limitations of traditional biometric authentication methods. Conventional biometric systems often rely on static fusion techniques, which may fail to adapt effectively to varying environmental conditions, noise, and modality reliability. Therefore, intelligent and adaptive fusion strategies are essential for improving recognition accuracy and robustness in security-critical applications.

Objective

This research aims to develop a Deep Reinforcement Learning (DRL)-based identity recognition framework that adaptively fuses facial, iris, and fingerprint biometric modalities to enhance authentication performance under challenging real-world conditions.

Methods

The proposed framework collects biometric data from publicly available datasets and controlled experimental settings. Preprocessing techniques, including image normalization, alignment, and noise reduction, are applied to improve data quality. Convolutional Neural Networks (CNNs) are employed to extract discriminative feature representations from facial, iris, and fingerprint images. A Modified Lévy Flight Distribution-enhanced Deep Deterministic Policy Gradient (MLFD-DDPG) algorithm is introduced to perform adaptive fusion by dynamically optimizing modality weights through reward-driven learning. The fused feature vectors are subsequently classified using a fully connected neural network with a softmax classifier.

Results

Experimental evaluations conducted in Python on benchmark biometric datasets demonstrate that the proposed MLFD-DDPG framework achieves a recognition accuracy of 98.73%, outperforming conventional static fusion approaches in robustness, adaptability, and recognition performance.

Conclusion

The proposed DRL-driven biometric system provides an intelligent, scalable, and highly reliable identity recognition solution suitable for advanced security and authentication applications.