Sheng Zhang

Dr. Sheng Zhang


Research Focus & Metrics

Academic Career Length

15Years (2012-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
7 February 2023 in Sensors
Verified
Journal Article
Open access
PDF file
JXJiandong Xu
Jiandong Xu
JPJiong Pan
Jiong Pan
TCTianrui Cui
Tianrui Cui
SZSheng Zhang
Sheng Zhang
Yi YangYi Yang
Yi Yang
TRTian-Ling Ren
Tian-Ling Ren
Human–Machine Interface (HMI) plays a key role in the interaction between people and machines, which allows people to easily and intuitively control the machine and immersively experience the virtual world of the meta-universe by virtual reality/augmented reality (VR/AR) technology. Currently, wearable skin-integrated tactile and force sensors are widely used in immersive human–machine interactions due to their ultra-thin, ultra-soft, conformal characteristics. In this paper, the recent progress of tactile and force sensors used in HMI are reviewed, including piezoresistive, capacitive, piezoelectric, triboelectric, and other sensors. Then, this paper discusses how to improve the performance of tactile and force sensors for HMI. Next, this paper summarizes the HMI for dexterous robotic manipulation and VR/AR applications. Finally, this paper summarizes and proposes the future development trend of HMI.

20 October 2016 in Sensors
Verified
Journal Article
Open access
PDF file
MWMinglin Wu
Minglin Wu
SZSheng Zhang
Sheng Zhang
YDYuhan Dong
Yuhan Dong
In this article, a novel driving behavior recognition system based on a specific physical model and motion sensory data is developed to promote traffic safety. Based on the theory of rigid body kinematics, we build a specific physical model to reveal the data change rule during the vehicle moving process. In this work, we adopt a nine-axis motion sensor including a three-axis accelerometer, a three-axis gyroscope and a three-axis magnetometer, and apply a Kalman filter for noise elimination and an adaptive time window for data extraction. Based on the feature extraction guided by the built physical model, various classifiers are accomplished to recognize different driving behaviors. Leveraging the system, normal driving behaviors (such as accelerating, braking, lane changing and turning with caution) and aggressive driving behaviors (such as accelerating, braking, lane changing and turning with a sudden) can be classified with a high accuracy of 93.25%. Compared with traditional driving behavior recognition methods using machine learning only, the proposed system possesses a solid theoretical basis, performs better and has good prospects.

7 June 2021 in Nano Select
Verified
Journal Article
Open access
PDF file
JXJiandong Xu
Jiandong Xu
RLRuisong Li
Ruisong Li
HXHaokai Xu
Haokai Xu
Yi YangYi Yang
Yi Yang
SZSheng Zhang
Sheng Zhang
TRTian-Ling Ren
Tian-Ling Ren
Glaucoma, a chronic optic neuropathy, is the leading cause of irreversible blindness in the world. Elevated intraocular pressure (IOP) has been considered to be the major contributor to glaucoma for a long time and is currently proved to be the only modifiable risk factor for the progression of optic neuropathy. IOP fluctuates throughout the day with a circadian rhythm change and is affected by body gesture changes. Moreover, the IOP spike usually occurs at night or in the early morning. Therefore, the current clinical practice of single and static measurements of IOP during office hours is not conducive to the early diagnosis and treatment of glaucoma. This review focuses on current advances in implantable and noninvasive IOP sensors for obtaining 24‐hour continuous IOP profiles. The content summarizes and classifies IOP sensors based on their working principles and provides representative examples of the sensors for IOP monitoring. Finally, the review further analyzes the challenges of current IOP sensors for clinical practice and puts forward the prospect of IOP sensors in the future.

1 January 2012
Verified
Book Chapter
Closed access
PDF file
MFMingjie Fang
Mingjie Fang
DLDongdong Li
Dongdong Li
JQJinguo Quan
Jinguo Quan
SZSheng Zhang
Sheng Zhang
XLXiaokang Lin
Xiaokang Lin

12 April 2018 in arXiv
Verified
Preprint
Open access
PDF file
YGYuechao Gao
Yuechao Gao
NLNianhong Liu
Nianhong Liu
SZSheng Zhang
Sheng Zhang
It is a challenging task to deploy computationally and memory intensive State-of-the-art deep neural networks (DNNs) on embedded systems with limited hardware resources and power budgets. Recently developed techniques like Deep Compression make it possible to fit large DNNs, such as AlexNet and VGGNet, fully in on-chip SRAM. But sparse networks compressed using existing encoding formats, like CSR or CSC, complex the computation at runtime due to their irregular memory access characteristics. In [1], we introduce a computation dataflow, stacked filters stationary dataflow (SFS), and a corresponding data encoding format, relative indexed compressed sparse filter format (CSF), to make the best of data sparsity, and simplify data handling at execution time. In this paper we present FPGA implementations of these methods. We implement several compact streaming fully connected (FC) and Convolutional (CONV) neural network processors to show their efficiency. Comparing with the state-of-the-art results [2,3,4], our methods achieve at least 2x improvement for computation efficiency per PE on most layers. Especially, our methods achieve 8x improvement on AlexNet layer CONV4 with 384 filters, and 11x improvement on VGG16 layer CONV5-3 with 512 filters.

20 May 2018
Verified
Conference Paper
Open access
PDF file
YGYuechao Gao
Yuechao Gao
NLNianhong Liu
Nianhong Liu
SZSheng Zhang
Sheng Zhang
It is a challenging task to deploy computationally and memory intensive State-of-the-art deep neural networks (DNNs) on embedded systems with limited hardware resources and power budgets. Recently developed techniques like Deep Compression make it possible to fit large DNNs, such as AlexNet and VGGNet, fully in on-chip SRAM. But sparse networks compressed using existing encoding formats, like CSR or CSC, complex the computation at runtime due to their irregular memory access characteristics. In [1], we introduce a computation dataflow, stacked filters stationary dataflow (SFS), and a corresponding data encoding format, relative indexed compressed sparse filter format (CSF), to make the best of data sparsity, and simplify data handling at execution time. In this paper we present FPGA implementations of these methods. We implement several compact streaming fully connected (FC) and Convolutional (CONV) neural network processors to show their efficiency. Comparing with the state-of-the-art results [2,3,4], our methods achieve at least 2× improvement for computation efficiency per PE on most layers. Especially, our methods achieve 8× improvement on AlexNet layer CONV4 with 384 filters, and 11× improvement on VGG16 layer CONV5-3 with 512 filters.