StarLKNet: Star Mixup with Large Kernel Networks for Palm Vein Identification

TL;DR

This work proposes StarLKNet, a large kernel convolution-based palm-vein identification network, with the Mixup approach, and designed LaKNet, a network leveraging large kernel convolution and gating mechanism in the domain of vein identification.

Abstract

As a representative of a new generation of biometrics, vein identification technology offers a high level of security and convenience.Convolutional neural networks (CNNs), a prominent class of deep learning architectures, have been extensively utilized for vein identification. Since their performance and robustness are limited by small \emph{Effective Receptive Fields} (\emph{e.g.}, 3$\times$3 kernels) and insufficient training samples, however, they are unable to extract global feature representations from vein images effectively. To address these issues, we propose \textbf{StarLKNet}, a large kernel convolution-based palm-vein identification network, with the Mixup approach.Our StarMix learns effectively the distribution of vein features to expand samples. To enable CNNs to capture comprehensive feature representations from palm-vein images, we explored the effect of convolutional kernel size on the performance of palm-vein identification networks and designed LaKNet, a network leveraging large kernel convolution and gating mechanism. In light of the current state of knowledge, this represents an inaugural instance of the deployment of a CNN with large kernels in the domain of vein identification. Extensive experiments were conducted to validate the performance of StarLKNet on two public palm-vein datasets. The results demonstrated that \textbf{StarMix} provided superior augmentation, and \textbf{LakNet} exhibited more stable performance gains compared to mainstream approaches, resulting in the highest identification accuracy and lowest identification error.

Figures (10)

• Figure 1: StarLKNet training workflow. After getting the mixing ratio, choose different mixing methods according to the threshold setting, get the mixed samples according to the different mixing methods, and then go through the encoder to get the final prediction and then calculate the loss to complete the training once.
• Figure 2: Left: Top-1 Accuracy($\uparrow$) using MixUp and StarMix in different models; Right: StarLKNet and ResNet18 fitting curves on the VERA220 dataset; StarLKNet fits faster and with higher classification accuracy.
• Figure 3: From left$\rightarrow$right are the visualizations of StarMask with different $\lambda$.
• Figure 4: The framework of the proposed LaKNet. $Upper:$ Denotes the main module of LaKNet, containing 1 Stem, 4 Stages, 3 Necks, and an FC layer. $Lower:$ a comprehensive illustration of the functions and specific operations within each module.
• Figure 5: (a). shows the ROC curves for different models on TJU600; (b). shows the ROC curves for different models using StarMix on TJU600; (c). shows the ROC curves for different models on VERA220; (d). shows the ROC curves for different models using StarMix on VERA220.