Yijun Mao

Yijun Mao

Auto-generated profile, based on publicly available information

Yijun Mao

College of Mathematics and Informatics, South China Agricultural University, Guangzhou 510642, China

Research Focus & Metrics

Academic Career Length

12Years (2015-Present)

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

Representative Publications

View All Publications
26 August 2020 in IEEE Transactions on Vehicular Technology
Verified
Journal Article
Closed access
DCDunjin Chen
Dunjin Chen
JWJian Weng
Jian Weng
Feiran HuangFeiran Huang
Feiran Huang
JZJian Zhou
Jian Zhou
YMYijun Mao
Yijun Mao
XLXiaodong Liu
Xiaodong Liu
Realtime localization and mapping in a cluttered and noisy indoor environment is a major problem in autonomous unmanned ground vehicle (UGV) navigation. This paper proposes a Heuristic Monte Carlo algorithm (HMCA) based on the Monte Carlo localization and Discrete Hough Transform (DHT) to build an autonomous navigation system. Specifically, a generalized map-processing method is first presented for the Hough Transform Algorithm (HTA), which can extract and cluster important map feature information and preserve the low computational complexity of real-time processing. Then a set of relative rotation angles and corresponding spatial displacements of the robot are obtained by constructing hough spatial energy spectrum correlation functions as global and local 2D occupancy grid maps (OGMs). Finally, the result is used as the guiding particle set of the Monte Carlo location algorithm. In a simulated scene and a real scene, the algorithm is tested several times, the adequacy of the algorithm is verified, and a map with good self-positioning performance is constructed.

19 April 2021 in IEEE Transactions on Neural Networks and Learning Systems
Verified
Journal Article
Closed access
HLHailiang Li
Hailiang Li
JWJian Weng
Jian Weng
YMYijun Mao
Yijun Mao
YWYonghua Wang
Yonghua Wang
YZYiju Zhan
Yiju Zhan
WGWanrong Gu
Wanrong Gu
Dropout is one of the most widely used methods to avoid overfitting neural networks. However, it rigidly and randomly activates neurons according to a fixed probability, which is not consistent with the activation mode of neurons in the human cerebral cortex. Inspired by gene theory and the activation mechanism of brain neurons, we propose a more intelligent adaptive dropout, in which a variational self-encoder (VAE) overlaps to an existing neural network to regularize its hidden neurons by adaptively setting activities to zero. Through alternating iterative training, the discarding probability of each hidden neuron can be learned according to the weights and thus effectively avoid the shortcomings of the standard dropout method. The experimental results in multiple data sets illustrate that this method can better suppress overfitting in various neural networks than can the standard dropout. Additionally, this adaptive dropout technique can reduce the number of neurons and improve training efficiency.

28 May 2024 in Journal of Magnetic Resonance Imaging
Verified
Journal Article
Closed access
YMYijun Mao
Yijun Mao
KIKiko Imahori
Kiko Imahori
WFWanxuan Fang
Wanxuan Fang
Hiroyuki SugimoriHiroyuki Sugimori
Hiroyuki Sugimori
SKShinji Kiuch
Shinji Kiuch
TKTamotsu Kamishima
Tamotsu Kamishima

1 March 2020 in Signal Processing
Verified
Journal Article
Closed access
QZQin Zhang
Qin Zhang
WLWei Lu
Wei Lu
THTao Huang
Tao Huang
SLShangjun Luo
Shangjun Luo
ZXZhaopeng Xu
Zhaopeng Xu
YMYijun Mao
Yijun Mao

11 July 2022 in Journal of Intelligent Information Systems
Verified
Journal Article
Closed access
PDF file
YHYichen He
Yichen He
YMYijun Mao
Yijun Mao
XXXianfen Xie
Xianfen Xie
WGWanrong Gu
Wanrong Gu

10 July 2020 in IEEE Access
Verified
Journal Article
Open access
WGWanrong Gu
Wanrong Gu
YMYijun Mao
Yijun Mao
YHYichen He
Yichen He
ZLZaoqing Liang
Zaoqing Liang
XXXianfen Xie
Xianfen Xie
WFWeijiang Fan
Weijiang Fan
With the development of computer vision recognition technology, more and more researchers apply this technology to the recognition of tumor images. But for cost reasons, many hospitals still use low-cost ultrasound and other cheap equipment, resulting in ambiguity, artifacts and many similar tumor noise areas images. Recent related studies have high precision in clear image recognition, such as face and number recognition in color images. However, they showed low accuracy and unstable results in ultrasonic image due to its fuzziness. The straight reason may be that many existing algorithms are not suitable for the fuzzy and noise image, and easily to misjudge the real objects and noise areas. In this paper, we proposed an approach based on R-CNN and RPN to distinguish the real objects from noise areas after data enhancement and morphological filtering with the optimization of a series of hyper-parameters and the combination of Color Doppler Flow Imaging (CDFI) blood flow signal model. We found that the key features of high-noise ultrasound images can be obtained quickly and accurately. From the experimental results, the accuracy and stability of our proposed method outperforms the state-of-the-art approaches on the real world thyroid tumor image data set.