Lei Liu

Dr. Lei Liu

Institute of Remote Sensing Satellite, China Academy of Space Technology, Beijing 100190, China

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
17 September 2025 in Remote Sensing
Verified
Journal Article
Open access
PDF file
JMJian Ma
Jian Ma
MBMingming Bian
Mingming Bian
FFFan Fan
Fan Fan
HKHui Kuang
Hui Kuang
LLLei Liu
Lei Liu
RZRunning Zhang
Running Zhang
Synthetic aperture radar (SAR), with its all-weather and all-day active imaging capability, has become indispensable for geoscientific analysis and socio-economic applications. Despite advances in deep learning–based object detection, the rapid and accurate detection of small objects in SAR imagery remains a major challenge due to their extremely limited pixel representation, blurred boundaries in dense distributions, and the imbalance of positive–negative samples during training. Recently, vision–language models such as Contrastive Language-Image Pre-Training (CLIP) have attracted widespread research interest for their powerful cross-modal semantic modeling capabilities. Nevertheless, their potential to guide precise localization and detection of small objects in SAR imagery has not yet been fully exploited. To overcome these limitations, we propose the CLIP-Driven Adaptive Tiny Object Detection Diffusion Network (CDATOD-Diff). This framework introduces a CLIP image–text encoding-guided dynamic sampling strategy that leverages cross-modal semantic priors to alleviate the scarcity of effective positive samples. Furthermore, a generative diffusion-based module reformulates the sampling process through iterative denoising, enhancing contextual awareness. To address regression instability, we design a Balanced Corner–IoU (BC-IoU) loss, which decouples corner localization from scale variation and reduces sensitivity to minor positional errors, thereby stabilizing bounding box predictions. Extensive experiments conducted on multiple SAR and optical remote sensing datasets demonstrate that CDATOD-Diff achieves state-of-the-art performance, delivering significant improvements in detection robustness and localization accuracy under challenging small-object scenarios with complex backgrounds and dense distributions.

9 August 2024 in Remote Sensing
Verified
Journal Article
Open access
PDF file
XTXiao Tan
Xiao Tan
ZYZhiwei Yang
Zhiwei Yang
HLHai xiang Li
Hai xiang Li
LLLei Liu
Lei Liu
XLXiaorui Li
Xiaorui Li
Bistatic radar exhibits spatial isomerism and diverse configurations, leading to unique clutter characteristics distinct from those of monostatic radar. The clutter rank serves as a pivotal indicator of clutter characteristics, enabling the quantification of clutter severity. Space-time adaptive processing (STAP) is a critical technique to detect moving targets, and clutter rank determines the number of independent and identically distributed (IID) training samples and the degree of freedom (DOF) for effective suppression of clutter that STAP requires. Therefore, the accurate estimation of clutter rank for bistatic radar can provide a crucial indicator for designing and constructing STAP processors, thereby facilitating fast and efficient clutter suppression in bistatic radar systems. This study is based on the idea that clutter rank is the number of prolate spheroidal wave function (PSWF) orthogonal bases utilized for approximating the clutter signal. Firstly, the challenge of utilizing PSWF orthogonal bases for approximating the clutter signal in bistatic radar is elucidated. This pertains to the fact that, unlike monostatic radar clutter, bistatic radar clutter is not capable of being expressed as a single-frequency signal. The clutter rank estimation for bistatic radar is thus derived as the frequency bandwidth estimation. Secondly, to achieve this estimation, the frequency distribution of each individual scattering unit is investigated, thereby determining their extending frequency broadening (EFB) as compared to that of single-frequency. Subsequently, the integral average of EFB across the entire range bin is computed, ultimately enabling the acquisition of bistatic radar’s frequency bandwidth. Finally, the estimation method is extended to non-side-looking mode and limited observation areas with pattern modulation. Simulation experiments confirm that our proposed method provides accurate clutter rank estimations, surpassing 99% proportions of large eigenvalues across various bistatic configurations, observation modes, and areas.

20 December 2012 in Journal of Electronics & Information Technology
Verified
Journal Article
Closed access
LLLei Liu
Lei Liu
XWXiao-Qing Wang
Xiao-Qing Wang
XYXiang-Zhen Yu
Xiang-Zhen Yu
JCJin-Song Chong
Jin-Song Chong
WHWen Hong
Wen Hong

19 July 2023 in Journal of Spatial Science
Verified
Journal Article
Closed access
JCJiajie Chen
Jiajie Chen
LLLei Liu
Lei Liu
XLXunchao Liu
Xunchao Liu
HMHongli Miao
Hongli Miao
QZQingjun Zhang
Qingjun Zhang

9 May 2023 in Remote Sensing
Verified
Journal Article
Open access
PDF file
XMXiangying Miao
Xiangying Miao
LLLei Liu
Lei Liu
HMHongli Miao
Hongli Miao
ZYZhonghao Yang
Zhonghao Yang
JWJing Wang
Jing Wang
QZQingjun Zhang
Qingjun Zhang
The baseline error is a primary error source of the wide-swath altimeter, directly related to the cross-track distance, and can lead to serious height errors at the swath’s outer edge. Cross-calibration using discrepancies with reference data can effectively estimate and correct the baseline error. However, building a reference surface that accurately describes the sea surface at the observation time is necessary to use this cross-correction method. The dynamic ocean environments where the sea surface structure changes over time are challenging. This paper proposes a method for constructing reference topography data (RTD) based on multi-source data products to correct the baseline error of the wide-swath altimeter. The effectiveness of the proposed method is evaluated using HYCOM ocean model data to assess the timeliness of the baseline error correction. The results demonstrate that using RTD at the observation time of the wide-swath altimeter can significantly correct the baseline error. The RMSE of the corrected sea surface height (SSH) in different regions is typically between 1~2 cm, except in some regions with strong currents where the RMSE is approximately 3~4 cm. However, the time interval between the RTD and the observation time of the wide-swath altimeter can affect the accuracy of the baseline error correction. The timeliness of this correction is influenced by the variability of SSH in different regions. In regions with relatively slow SSH changes near the equator, the effective time based on HYRTD and MORTD can basically reach more than 7 days. In regions where the SSH changes more rapidly, the correction result may no longer be reliable in only 1~3 days.

21 August 2025 in Remote Sensing
Verified
Journal Article
Open access
PDF file
YWYanping Wang
Yanping Wang
YJYunzhen Jia
Yunzhen Jia
WSWenjie Shen
Wenjie Shen
YLYun Lin
Yun Lin
YLYang Li
Yang Li
QZQingjun Zhang
Qingjun Zhang
Air moving-target detection is crucial for national defense, civil aviation, and airspace supervision. Spaceborne synthetic aperture radar (SAR) provides high-resolution, continuous observations for this task, but faces challenges including target attitude variation-induced weak signals and Doppler defocusing from targets’ high-speed motion, which hinder target-background separation. To address this, we propose a novel method combining sub-aperture segmentation with GoDec+ low-rank decomposition to enhance signal-to-noise ratio and suppress defocusing. Critically, ADS-B flight data is integrated as ground truth for spatio-temporal validation. Experiments using Sentinel-1 SM mode SLC imagery across farmland, forest, and mountainous regions confirm the method’s effectiveness and robustness in real airspace scenarios.