Xingwei Li

Xingwei Li

Auto-generated profile, based on publicly available information

Xingwei Li

Information Engineering University, Zhengzhou, 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
23 April 2019 in IEEE Access
Verified
Journal Article
Open access
XLXingwei Li
Xingwei Li
ZSZheng Shan
Zheng Shan
FLFudong Liu
Fudong Liu
YCYihang Chen
Yihang Chen
YHYifan Hou
Yifan Hou
Packing is the most common malware obfuscation technique employed to evade detection by anti-virus systems. With the explosive growth of packed malware, packer identification has become increasingly important in current cybersecurity. Most prior researches lack resilience since they neglect the significance of preserving the semantic integrity for pursuing efficiency. Therefore, in this paper, we propose a novel approach based on consistently-executing graph (CEG) mining, which can maximise semantics and facilitate the execution intents of executables for identification. We elaborate on the concepts of consistent execution and represent malware packers as structured CEGs. Then we employ the bipartite matching algorithm built on a designed block proximity metric to extract the critical matching subgraphs. Weisfeiler-Lehman shortest path graph kernel is adopted for pairwise comparison to make the prediction. We have evaluated the efficacy extensively with several experiments on large-scale datasets containing manually-packed benign apps, wild-packed malware and wild-unpacked malware. Promising results demonstrate that our approach is valid and practical when applied to real-world packed malware analysis.

22 April 2025 in IEEE Transactions on Information Forensics and Security
Verified
Journal Article
Closed access
XLXingwei Li
Xingwei Li
YKYan Kang
Yan Kang
CWChenggang Wu
Chenggang Wu
DLDanjun Liu
Danjun Liu
JWJiming Wang
Jiming Wang
QWQiang Wei
Qiang Wei
The Linux kernel remains vulnerable to numerous bugs, with approximately 65% detected by Syzkaller lacking Proof-of-Concept (PoC), hampering risk mitigation efforts. These bugs, termed irreproducible kernel bugs, highlight the challenge of statefulness issue-related irreproducibility in kernel fuzzing, which is an open research without definitive solutions. Our investigation reveals that suboptimal seed quality distribution in fuzzing is the root obstacle preventing effective tracking of the states leading to crashes. Inspired by this insight, we introduce Reverse Fuzzing (RF), an innovative approach that infers hard-to- reach states by continuously reverse-oriented deriving from subsequently encountered bridge states to increase reproduction probability. RF differentiates between the “trigger” seed, which directly causes crashes, and “activator” seeds, which establish the necessary preconditions, prioritizing exploration around trigger while simultaneously regenerating and maintaining activators during fuzzing, which effectively facilitate to restructure such elusive states from “yesterday”. We implement YOME, a prototype leveraging RF to strike a balance between fuzzing efficiency and effectiveness through customized scheduling and mutation strategies, armed with a refinement mechanism to improve seed quality distribution. Our evaluations validate that YOME reproduce 110% more bugs than previous kernel fuzzers and demonstrate its practicality in real-world scenarios. YOME generated 125 PoCs (30.1% of the total) and uncovered 23 unique bugs, with 40 confirmed and 5 assigned CVEs.

19 November 2025
Verified
Conference Paper
Open access
YSYue Sun
Yue Sun
YKYan Kang
Yan Kang
CWChenggang Wu
Chenggang Wu
KLKangjie Lu
Kangjie Lu
JWJiming Wang
Jiming Wang
ZWZhe Wang
Zhe Wang

1 October 2015
Verified
Conference Paper
Closed access
JGJinlu Gao
Jinlu Gao
YFYuqiang Fang
Yuqiang Fang
XLXingwei Li
Xingwei Li
With the rising of intelligent vehicle technologies, traffic sign recognition become an essential problem in computer vision. Focusing on the traffic sign recognition under real-world scenario, this paper aims to develop novel local feature representation to improve the traffic sign recognition performance. Especially, with the local histogram feature as a basic unit, a novel histogram intersection kernel based dictionary learning method is proposed for feature quantization. Then a fast feature encoding approach based on look-up table is induced to improve the calculation effectiveness. The proposed recognition method achieves high performance on several off-line traffic sign databases, and has also been extended to recognize traffic sign in real-world videos. Extensive experiments have demonstrated the effectiveness of new method.

1 January 2021
Verified
Conference Paper
Open access
JJJiating Jin
Jiating Jin
XLXingwei Li
Xingwei Li
SGShaojie Guan
Shaojie Guan
XLXinlong Li
Xinlong Li
XZXianfa Zhao
Xianfa Zhao
Multi-object tracking is an important problem in the field of computer vision. However, it has to face a lot of challenges, such as missed targets, occlusion, similar appearance, and the complex movement of objects. In order to cope with the above issues, we propose a multi-object tracking framework with Long Short-Term Memory (LSTM). Firstly, in order to handle the occlusion and complex motion of objects in the tracking process, a LSTM motion model is established. Secondly, a two-stage tracking framework is applied to deal with the missed targets and ID switches, in which trajectories are clustered and optimized after solving the network flow model. Experimental results on the public MOT16 dataset demonstrate that our algorithm has a promising performance compared with other state-of-art methods.

1 January 2021
Verified
Conference Paper
Open access
XLXinlong Li
Xinlong Li
XLXingwei Li
Xingwei Li
JJJiating Jin
Jiating Jin
SGShaojie Guan
Shaojie Guan
YGYizhi Ge
Yizhi Ge
Few shot learning has become a research focus in the field of computer vision. Few shot detection method has a very crucial value in practice. But so far, only few results have been achieved in this field, and more research is required. In this paper, we proposed a few shot object detection method, which consists of two subnets. The first subnet is the object detection subnet, which is only responsible for the object positioning. We adopted an anchor box clustering method to produce the proper initial size of anchor boxes. Furthermore, a general class classification method was proposed to make the model learn the common features of different classes. Therefore, it could detect unseen classes when testing. The second subnet is the few shot classification subnet, which is designed to classify the detected objects. Images were preprocessed before input to the few shot detection subnet to remove redundant information. Finally, end-to-end experiments were designed on the ImageNet LOC dataset. Our method reached a high mAP of 80.76% in 5-way-10-shot experiments and significantly outperformed the benchmark method.