Research On Iris Recognition Algorithm Based On Deep Neural Network

Biometric recognition technology,compared to traditional identity verification methods,has several advantages such as being difficult to forge,convenient and fast,high accuracy,and increased security.With the growing demand for information protection and security,iris recognition technology has gradually gained the attention of researchers and has been widely applied in fields such as public safety,judiciary,and education.The most critical step in the iris recognition process is iris feature extraction and matching,and the quality of iris recognition algorithms directly affects the performance of iris recognition systems.Although iris recognition algorithms based on deep neural networks have made significant progress,they still face some challenges.Existing research often lacks a welldesigned convolutional operation in the feature extraction stage,which can lead to inefficient feature extraction.Additionally,there’s room for improvement in utilizing the extracted features effectively during the feature matching stage.To address the aforementioned shortcomings,this article proposes an iris recognition algorithm based on deep neural networks,with the main tasks outlined as follows:1.Iris Feature Extraction Phase: This article introduces the EfficientIrisNet model,which employs large convolutional kernels to extract shallow-level features,reducing interference caused by local noise and distortions in iris images.It incorporates a multi-branch convolutional structure(Multi-branch convolution,MBC)to fuse multi-scale deep features,enhancing feature richness.Additionally,a Selective Kernel Attention Mechanism(SKAM)is introduced to adaptively adjust the fusion weights of different branches in MBC.Iris Feature Matching Phase: First,cosine distance is utilized to calculate the value similarity between iris feature vectors.Next,the first-order temporal correlation coefficient is employed to compute behavioral similarity.Finally,a distance metric is applied to iris features using both value and behavioral similarity to better utilize the iris features extracted in the feature extraction phase,thus enhancing the robustness of the feature matching method.2.In the experiments,this paper selected six open-source iris datasets including JLU-6.0and JLU-7.0,with evaluation metrics such as recognition rate and equal error rate.The rationality of the model’s design was demonstrated through empirical experiments,conducted both within each dataset and across different datasets.The results obtained were promising across multiple datasets,confirming the effectiveness of this paper’s algorithm.3.Deployment of the Iris Recognition System: Based on the iris recognition algorithm described above,an iris recognition system was designed and implemented.The system comprises five modules: the acquisition module,interface interaction module,storage module,control module,and iris recognition module.The most critical component is the iris recognition module,which is based on the iris feature matching and recognition algorithm proposed in this paper.It can be applied to a broader range of iris recognition scenarios and holds practical value.