Improving Iris Recognition Systems with Transfer Learning and Pretrained CNN Models

Conventional iris identification systems have failed to manage environmental variables and changes in iris patterns, sometimes depending on handmade characteristics that may lack the delicate details required for reliable identification. The paper offers a revolutionary iris identification system based on deep learning, especially pre-trained Convolutional Neural Networks (CNNs) such as ResNet and VGG, which are fine-tuned utilizing transfer learning methods. The system is trained and verified using a variety of iris datasets, taking into account illumination fluctuations, occlusions, and other environmental conditions. Using these pre-trained CNN models, the proposed system intends to dramatically improve the accuracy and reliability of iris detection, even under difficult settings. The proposed system attains significant gains in recognition accuracy, even under challenging circumstances, thorough preprocessing, transfer learning, fine-tuning, and data augmentation. Testing and validation show how effective the proposed system is; it has better performance metrics than existing systems, with 98.5% accuracy, 97.8% precision, 98.9% recall, and 98.3% F1-score, which are all higher than those of existing systems. The methodology also guarantees scalability and computing efficiency, confirming the proposed system's potential for accurate and dependable iris detection in practical settings. The reliability and efficiency of iris recognition technology have been significantly improved by these advances.