A New Image Recognition Combining Transfer Learning Algorithm and MobileNet V2 Model for Palm Vein Recognition

Among human biometrics, palm vein recognition is characterised by high security and privacy as an internal human biometric identification technology. Despite the breakthroughs associated with palm vein technology, there are various factors such as lighting variations, ambient temperature, and angular hand position that make it difficult to extract palm vein features from the original image. Also, due to the lack of sufficient palm vein datasets, the models will not have enough samples to differentiate the features and will easily lead to overfitting of the data. To address the above issues, we propose a MobileNet V2 convolutional neural network model combining a transfer learning algorithm named SEM. It consists of proposing a region of interest (ROI) segmentation method that contains more vein information, and transferring the pre-trained model weights from the palm print dataset to the palm vein recognition domain. Furthermore, the SENet module is embedded into the MobileNet V2 network to enhance the perceptual field of the network, and the important features are reinforced to improve the accuracy. Finally, the proposed SEM model is trained on the dataset, and our proposed method has a very high accuracy and high robustness, with an accuracy of 99.35%.