Yilun Chen

ORCID: 0000-0003-0618-3621
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Remote-Sensing Image Classification
  • Synthetic Aperture Radar (SAR) Applications and Techniques
  • Autonomous Vehicle Technology and Safety
  • Traffic control and management
  • Advanced SAR Imaging Techniques
  • Automated Road and Building Extraction
  • Soil Moisture and Remote Sensing
  • Underwater Acoustics Research
  • Reinforcement Learning in Robotics
  • GNSS positioning and interference
  • Vehicle Dynamics and Control Systems
  • Geophysics and Gravity Measurements
  • Image and Object Detection Techniques
  • Anomaly Detection Techniques and Applications
  • Advanced Memory and Neural Computing
  • Ionosphere and magnetosphere dynamics
  • Human Pose and Action Recognition
  • Power Systems and Technologies
  • Satellite Image Processing and Photogrammetry
  • Soil Geostatistics and Mapping
  • Advanced Image Fusion Techniques
  • Inertial Sensor and Navigation
  • Smart Grid and Power Systems
  • Blind Source Separation Techniques
  • Data Management and Algorithms

Beijing Satellite Navigation Center
2021-2024

Tsinghua University
2006-2024

Carnegie Mellon University
2019

University of Michigan
2008-2010

We propose a new approach to adaptive system identification when the model is sparse. The applies ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> relaxation, common in compressive sensing, improve performance of LMS-type methods. This results two algorithms, zero-attracting LMS (ZA-LMS) and reweighted (RZA-LMS). ZA-LMS derived via combining norm penalty on coefficients into quadratic cost function, which generates zero attractor...

10.1109/icassp.2009.4960286 article EN IEEE International Conference on Acoustics Speech and Signal Processing 2009-04-01

A pretest approach based on the complex Wishart distribution in polarimetric synthetic aperture radar (POLSAR) speckle filtering is proposed this paper. The main principle to select homogeneous pixels a large-scale area process, which called pretesting. To preserve details and fine structures while despeckling, are selected by comparing their 3 × neighboring windows. test statistic used decide selection of pixels. Speckle processed summing up with weights according values statistics....

10.1109/tgrs.2010.2087763 article EN IEEE Transactions on Geoscience and Remote Sensing 2010-12-11

In this letter, polarization cross-entropy is introduced based on the eigendecomposition of polarimetric coherence matrix. Then, new parameter employed for ship detection. From experimental results, it derived that distribution in ocean regions can be well approximated by a generalized exponential distribution. constant-false-alarm-rate ship-detection method proposed cross-entropy. Using authors demonstrate effectiveness

10.1109/lgrs.2009.2024224 article EN IEEE Geoscience and Remote Sensing Letters 2009-07-30

Performing safe and efficient lane changes is a crucial feature for creating fully autonomous vehicles. Recent advances have demonstrated successful following behavior using deep reinforcement learning, yet the interactions with other vehicles on road are rarely considered. In this paper, we design hierarchical Deep Reinforcement Learning (DRL) algorithm to learn change behaviors in dense traffic. By breaking down overall sub-policies, faster safer actions can be learned. We also apply...

10.1109/cvprw.2019.00172 article EN 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) 2019-06-01

Trajectory representation learning plays a pivotal role in supporting various downstream tasks, such as travel time estimation, trajectory classification and Top-k similar search. Traditional methods order to filter the noise GPS trajectories tend focus on routing-based simplify trajectories. However, these approaches ignore motion details contained data, limiting capability of learning. To fill this gap, we propose novel framework that is Jointly G PS Route Modeling based self-supervised...

10.1145/3589334.3645644 article EN Proceedings of the ACM Web Conference 2022 2024-05-08

Performing safe and efficient lane changes is a crucial feature for creating fully autonomous vehicles. Recent advances have demonstrated successful following behavior using deep reinforcement learning, yet the interactions with other vehicles on-road are rarely considered. In this paper, we design hierarchical Deep Reinforcement Learning (DRL) algorithm to learn change behaviors in dense traffic. By breaking down overall sub-policies, faster safer actions can be learned. We also apply...

10.1109/iros40897.2019.8968565 article EN 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2019-11-01

Recently, the use of linear features for processing remote-sensing images has shown its importance in applications. Unfortunately, traditional feature detection methods rely heavily on image's local information which makes them vulnerable to presence noise image. This problem becomes particularly difficult synthetic aperture radar (SAR) image applications where SAR are corrupted by speckle noise. In order overcome this problem, we propose a novel method that processes polarimetric (Pol-SAR)...

10.1109/tgrs.2010.2081373 article EN IEEE Transactions on Geoscience and Remote Sensing 2010-11-23

A novel method is presented to detect roads in synthetic aperture radar (SAR) images. multi-segmented poly-line model introduced provide a more accurate description of the road as well ensure curve's smoothness level. We then solve detection problem using Bayesian tracking theory, where particle filtering algorithm adopted simple and consistent framework. The effectiveness robustness proposed demonstrated by experimental results.

10.1109/icip.2006.312855 article EN International Conference on Image Processing 2006-10-01

In this paper, a new method for supervised classification of terrain types in polarimetric Synthetic Aperture Radar (Pol-SAR) images is proposed. This technique combination the texture and maximum likelihood based on complex Wishart distribution covariance matrix. The features are first extracted from span image co-occurrence matrices; then classifier combines with distance measure information to obtain results. Using NASA/JPL AIRSAR image, effectiveness proposed demonstrated.

10.1109/radar.2010.5494572 article EN IEEE Radar Conference 2010-01-01

Abstract A new method is proposed for joint detection of roads in multifrequency synthetic aperture radar (SAR) images. First, a multisegmented polyline model was introduced to provide more accurate description road curve. Then, the SAR images were extracted Bayesian tracking framework, and particle filtering algorithm used implement tracking. Finally, based on maximum likelihood (ML) criterion determine optimal weights particles. Using data from National Aeronautics Space Administration Jet...

10.1080/01431160903252319 article EN International Journal of Remote Sensing 2010-02-18

In this paper, a novel scheme for supervised classification problem of Polarimetric SAR images is proposed, which based on Adaboost. Compared to traditional classifiers such as complex Wishart distribution maximum likelihood classifier or Neural Network classifier, the proposed method more robust and flexible. Different features parameters extracted from data could be adopted into quantitative analysis significance each parameter achieved. Experiment results demonstrated effectiveness scheme.

10.1109/icosp.2008.4697633 article EN 2008-10-01

Pseudorange bias refers to the receiver-dependent and satellite-dependent constant in pseudorange resulting from nonideal characteristics of a signal. The impact on high-precision satellite navigation services has long been ignored. This paper proposes calibration method for two collocated receivers. Then, we calibrate biases types receivers at monitoring station within China evaluate their services: BeiDou Navigation Satellite System 3 (BDS-3) dual-frequency augmentation precise point...

10.3390/rs16163022 article EN cc-by Remote Sensing 2024-08-17

A new method for SAR image segmentation is proposed in this paper. Region can be achieved by contour tracking, and we use the general Bayesian tracking framework to solve problem. Due non-linearity of problem non-Gaussian noise image, Monte Carlo based particle filtering algorithm adopted obtain optimal solution. Based on framework, a filter region images. In method, each assigned linear segment with specific location direction. The response local edge detector used calculate weight while...

10.1109/icosp.2006.345673 article EN 2006-01-01

Through the deployment of monitoring receivers in different areas, continuous BeiDou RDSS service quality is realized. Service parameters include success rate/service delay/location deviation/timing accuracy, etc. The receiver records time application and total number applications, reception services received, returns this information to analysis center. However, due long-term operation instability ground network, there will be problems such as data interruption/change counters. This paper...

10.1109/iaeac50856.2021.9390725 article EN 2022 IEEE 6th Advanced Information Technology, Electronic and Automation Control Conference (IAEAC ) 2021-03-12

In this paper, a new approach to speckle filtering of synthetic aperture radar (SAR) data is presented. We define parameter space consisting two orthogonal subspaces - the signal subspace and noise subspace. Then, full polarimetric information from obtained after filtering. Moreover, edges different kinds targets are preserved. Using SAR data, effectiveness proposed method not only validate by standard-deviation-to mean ratio, but also target classification.

10.1109/mape.2005.1617928 article EN IEEE International Symposium on Microwave, Antenna, Propagation and EMC Technologies for Wireless Communications 2006-04-28
Coming Soon ...