Hardware-Friendly Palm Vein Recognition Algorithm Using Hybrid Compressed Convolutional Neural Network

With the rapid development of information technology, the demand for identity authentication is continuously increasing. Compared to other biometric systems, palm vein recognition offers several advantages, including a high level of security, difficulty to forge, low risk of damage, and high user acceptance. In recent years, deep learning has been increasingly applied to palmar vein recognition; however, their large network size makes them challenging to deploy on edge devices. To address this issue, this paper proposes a hardwarefriendly palm vein recognition algorithm based on MobileNetV3. First, we designed a palm vein ROI extraction algorithm using edge-centroid distance detection, achieving a 100% success rate in ROI extraction on the PolyU dataset. Subsequently, the input channels and final classifications of MobileNetV3-Small were reduced, and through experimental analysis, the width coefficient <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\alpha$</tex> was established at 0.5. This resulted in an initial palm vein recognition algorithm with an accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 9. 5 \%}$</tex> on the PolyU dataset. Furthermore, a hybrid compression method combining L2-norm filter pruning and quantization was applied. This approach further compressed the model to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 1 7 M}$</tex> while maintaining an accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 8. 2 9 \%}$</tex>. Compared with similar designs in recent years, it maintains higher accuracy and greatly reduces the scale of model parameters, which is conducive to deployment on edge devices.