Improved Lightweight Convolutional Neural Network for Finger Vein Recognition System.

Area of Science:

• Bioengineering
• Computer Vision
• Biometrics

Background:

• Computer vision (CV) and convolutional neural networks (CNNs) excel at feature extraction in bioengineering.
• Traditional CNNs struggle with translation in finger-vein imaging, hindering real-time, accurate recognition.
• CNNs possess numerous parameters and lack inherent shift-invariant features, limiting their practical application.

Purpose of the Study:

• To develop an improved lightweight convolutional neural network (ILCNN) for enhanced finger vein recognition.
• To address the limitations of existing CNNs in handling translation and parameter efficiency for biometrics.
• To improve the accuracy and robustness of finger vein identification systems in diverse environments.

Main Methods:

• Introduction of an improved lightweight convolutional neural network (ILCNN) incorporating a diverse branch block (DBB).
• Integration of adaptive polyphase sampling (APS) and a coordinate attention mechanism (CoAM) to boost feature identification.
• Evaluation using the finger-vein by university sains malaysia (FV-USM) and PLUSVein dorsal-palmar finger-vein (PLUSVein-FV3) public databases.

Main Results:

• The ILCNN model achieved high recognition accuracy rates of 99.82% on FV-USM and 95.90% on PLUSVein-FV3.
• The proposed model utilizes a significantly reduced parameter count of only 1.23 million.
• Comparative analysis demonstrated superior performance of the ILCNN over previous finger vein recognition methods.

Conclusions:

• The developed ILCNN offers a highly accurate and parameter-efficient solution for finger vein recognition.
• The integration of DBB, APS, and CoAM effectively enhances the model's ability to capture discriminative finger vein features.
• This research presents a promising advancement for real-time and reliable biometric identification systems.