RFEUnet Receptive Field Expansion for Enhanced Iris Segmentation Accuracy and Efficiency

Iris segmentation a critical step in iris recognition requires high accuracy and efficiency especially in real-world applications This paper presents RFEUnet a novel lightweight model designed to boost iris segmentation performance By integrating dilated convolutions CSP blocks and a coordinate attention mechanism RFEUnet expands the receptive field and strengthens feature extraction capabilities Experiments on the UBIRIS V2 dataset demonstrate that RFEUnet achieves superior segmentation accuracy with 9312% mIOU and faster processing speed at 26 FPS compared to the baseline DenseUnet model Our findings indicate RFEUnet is a promising solution for deploying iris recognition systems in resource constrained environments The source code is publicly available at GitHub https://github.com/ljd020403-prog/-.git with a persistent DOI:105281/zenodo.17562923. Experiments can be fully replicated using the provided code and UBIRIS V2 dataset