Prototype of Low Complexity CNN Hardware Accelerator with FPGA-based PYNQ Platform for Dual-Mode Biometrics Recognition
Yu-Hsiang Chen, Chih‐Peng Fan, KyungHi Chang
Abstract
In this study, the effective low-complexity convolutional neural network (CNN) inference network is implemented by the FPGA-based hardware accelerator for dual-mode biometric authentications. After the pre-processing processes, the selected datasets, which include the finger vein images and eye images with partial iris and sclera zones, are used for training and testing the LeNet-5 based CNN lite model. Then the lite CNN classifier will be rapidly prototyped by FPGA for hardware acceleration. By tests, the proposed lite CNN model achieves the recognition accuracy up to 97% with the ROI-based eye images. Besides, the proposed model achieves the recognition accuracy up to 95% with the finger vein images. Compared with the pure software based implementation, the proposed lite CNN hardware acceleration design provides the same recognition accuracy, and the inferential calculations are accelerated by about 50 times on the PYNQ FPGA platform.