Ke Li

Ke Li

Auto-generated profile, based on publicly available information

Ke Li

School of Computer Science and Technology, Xinjiang Normal University, Urumqi 830054, China

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

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

Representative Publications

View All Publications
10 March 2023 in Multimedia Tools and Applications
Verified
Journal Article
Closed access
PDF file
KLKe Li
Ke Li
Zhandong LiuZhandong Liu
Zhandong Liu

22 August 2025 in Applied Sciences
Verified
Journal Article
Open access
PDF file
Zhandong LiuZhandong Liu
Zhandong Liu
SRSong Ruixia
Song Ruixia
KLKe Li
Ke Li
YLYong Li
Yong Li
Scene text understanding, serving as a cornerstone technology for autonomous navigation, document digitization, and accessibility tools, has witnessed a paradigm shift from traditional methods relying on handcrafted features and multi-stage processing pipelines to contemporary deep learning frameworks capable of learning hierarchical representations directly from raw image inputs. This survey distinctly categorizes modern scene text recognition (STR) methodologies into three principal paradigms: two-stage detection frameworks that employ region proposal networks for precise text localization, single-stage detectors designed to optimize computational efficiency, and specialized architectures tailored to handle arbitrarily shaped text through geometric-aware modeling techniques. Concurrently, an in-depth analysis of text recognition paradigms elucidates the evolutionary trajectory from connectionist temporal classification (CTC) and sequence-to-sequence models to transformer-based architectures, which excel in contextual modeling and demonstrate superior performance. In contrast to prior surveys, this work uniquely emphasizes several key differences and contributions. Firstly, it provides a comprehensive and systematic taxonomy of STR methods, explicitly highlighting the trade-offs between detection accuracy, computational efficiency, and geometric adaptability across different paradigms. Secondly, it delves into the nuances of text recognition, illustrating how transformer-based models have revolutionized the field by capturing long-range dependencies and contextual information, thereby addressing challenges in recognizing complex text layouts and multilingual scripts. Furthermore, the survey pioneers the exploration of critical research frontiers, such as multilingual text adaptation, enhancing model robustness against environmental variations (e.g., lighting conditions, occlusions), and devising data-efficient learning strategies to mitigate the dependency on large-scale annotated datasets. By synthesizing insights from technical advancements across 28 benchmark datasets and standardized evaluation protocols, this study offers researchers a holistic perspective on the current state-of-the-art, persistent challenges, and promising avenues for future research, with the ultimate goal of achieving human-level scene text comprehension.

17 February 2023
Verified
Conference Paper
Open access
KLKe Li
Ke Li
Zhandong LiuZhandong Liu
Zhandong Liu
NDNan Ding
Nan Ding
RJRonghua Jiang
Ronghua Jiang
With the rapid development of our society and economy, traffic congestion caused by the increase in vehicles is an important challenge we are facing today. Traffic control, as an integral part of intelligent transport systems (ITS), plays a huge role in traffic control areas. A good signal control method can greatly improve the traffic capacity of vehicles on the road. In this paper, based on the traditional DQN algorithm, we propose a gradient algorithm based on meta-learning to promote the adjustment of the training parameters of the DQN algorithm and add an attention mechanism in the depth neural network to aggregate the effective features. Finally, we adjust the loss function to obtain a more effective DQN algorithm for road signal control. Simulation experiments conducted by SUMO simulation software prove the effectiveness of our method.

24 May 2024
Verified
Conference Paper
Open access
GLGang Liu
Gang Liu
Zhandong LiuZhandong Liu
Zhandong Liu
YRYuan Ren
Yuan Ren
HLHaifang Li
Haifang Li
NDNan Ding
Nan Ding
KLKe Li
Ke Li
In recent research, shift convolution has been discussed for implementing large kernel convolutions in convolutional neural networks using small kernels, while maintaining hardware friendliness. Furthermore, real time end-to-end models are known for their impressive speed and accuracy.In this paper, we introduce the Shift-ConvNet and design a plug-and-play C3SC module by combining it with the C3 module from YOLOv5. Previous studies have mentioned that model perception extension is a crucial task, and we enhance the perceptual capabilities by using the decoder from RT-DETR as the detection head and adding a layer on top. In addition, we use a novel dynamic upsampling operator, Dysample, which excels in multiple dense prediction tasks. Combining the above elements, we propose the CSCNet network as a neck structure. Finally, attention modules effectively support feature extraction and enhancement. Using two excellent bounding box regression losses, MPD IoU and Focaler IoU, as benchmarks, we designed a new loss function, Focaler-MPD IoU. Experimental results show that our method significantly improves speed and accuracy compared to the baseline models, achieving a 7.5% increase in AP value and a 4.9% increase in AP50 value on the CoCo val2017 dataset.