Search

Recent times have witnessed significant progress in deep learning-based finger vein pattern extraction methods, but two unavoidable issues still remain to be addressed. One is that the model trained on a single finger vein dataset shows poor generalizability, and the model performance is limited by the image quality of the single dataset; the other is that it is hard for the deep model to extract real-time finger vein patterns because of its large number of parameters and poor real-time performance. To address the aforementioned issues, we propose a novel lightweight domain-adaptive segmentation framework (Lite-HDNet) that learns a generic representation of different domains to improve the extraction of finger vein patterns. We propose a multi-domain feature knowledge transfer strategy and a domain migration loss converter to enable the trunk network to learn the robust representations of different finger vein datasets as well as to compensate for the heterogeneity between them. In the proposed framework, two lightweight segmentation networks are designed as the trunk branch and the auxiliary branch to achieve real-time extraction of finger vein patterns. Our approach has been extensively tested on four finger vein datasets available to the public, and the results show that our Lite-HDNet not only improves segmentation performance on all datasets but also effectively reduces heterogeneity between different domains. In addition, we also validated the real-time performance of Lite-HDNet on NVIDIA embedded terminals, proving the outperformance of our approach compared with previous lightweight segmentation networks.

Monte Carlo simulation is a precise method to model light propagation in bio-tissues and has been considered the golden standard to estimate the result of other computation methods. But the huge computation burden limited the application. In this paper, we propose a parallel computing model using graphic card to accelerate the Monte Carlo simulation in 3-D voxelized media with the consideration of internal refraction. Optimization of the parallel mode is made by using segmentations and offered an extra boost of simulation speed. The acceleration efficiency affecting factors are investigated and the acceleration rate of the five segmented model is 32.6 times higher than non-GPU model and 1.66 times higher than non-optimized model for a real human head 3-D structure simulation.

By leveraging the 5G enabled V2X networks, the vehicles connected by cellular base-stations can support a wide variety of computation-intensive services. In order to solve the arisen challenges in end-to-end low-latency transmission and backhaul resources, mobile edge computing (MEC) is now regarded as a promising paradigm for 5G-V2X communications. Considering the importance of both reliability and delay in vehicle communication, this article innovatively envisions a joint computation and URLLC resource allocation strategy for collaborative MEC assisted cellular-V2X networks and formulate a jointly power consumption optimization problem while guaranteeing the network stability. To solve this NP hard problem, we decouple it into two sub-problems: URLLC resource allocation for multi-cells to multi-vehicles and computation resource decisions among local vehicle, serving MEC server and collaborative MEC server. Secondly, non-cooperative game and bipartite graph are introduced to reduce the inter-cell interference and decide the channel allocation, which aims to maximize the throughput with a guarantee of reliability in URLLC V2X communication. Then, an online Lyapunov optimization method is proposed to solve computation resource allocation to get a trade-off between the average weighted power consumption and delay where CPU frequency are calculated using Gauss-Seidel method. Finally, the simulation results demonstrate that our proposed strategy can get better trade-off performance among power consumption, overflow probability and execution delay than the one based on centralized MEC assisted V2X.

A powered exoskeleton on reasonable design can effectively reduce the metabolic cost during walking. In order to explore the metabolism and muscle strength with powered exoskeleton assisted on joints under different states, we involved seven healthy subjects to walk on level and slope of 5° with different loads (0, 10, 20 and 30 kg), based on which we simulated the states of walking with different assistant torque on joints to analyze the metabolic cost and muscle force. The results demonstrated that the metabolism decreased more with the increase of hip extension assistance on flat and it had a more significant reduction under greater driving moment of hip flexion, hip extension, knee extension and ankle plantarflexion on slope. Both ankle plantarflexion and hip extension devices reduced more energy consumption than other powered devices, between which hip extension assistance increased the metabolic reduction rate significantly, especially on slope. Each kind of powered exoskeleton reduced the strengths of agonistic muscles assisted. In addition, hip extension device greatly reduced the strength of vastus intermedius and ankle plantarflexion device reduced the strengths of rectus femoris and vastus intermedius concurrently. The results predicted metabolism and muscle strength under different assistances in various states, providing advisement for drive design theoretically.