Liefa Liao

Dr. Liefa Liao


Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

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

Representative Publications

View All Publications
27 January 2025 in Sensors
Verified
Journal Article
Open access
PDF file
LLLiefa Liao
Liefa Liao
CSChao Song
Chao Song
SWShouluan Wu
Shouluan Wu
FJFU Jianglong
FU Jianglong
To address challenges like manual processes, complicated detection methods, high false alarm rates, and frequent errors in identifying defects on steel surfaces, this research presents an innovative detection system, YOLOv10n-SFDC. The study focuses on the complex dependencies between parameters used for defect detection, particularly the interplay between feature extraction, fusion, and bounding box regression, which often leads to inefficiencies in traditional methods. YOLOv10n-SFDC incorporates advanced elements such as the DualConv module, SlimFusionCSP module, and Shape-IoU loss function, improving feature extraction, fusion, and bounding box regression to enhance accuracy. Testing on the NEU-DET dataset shows that YOLOv10n-SFDC achieves a mean average precision (mAP) of 85.5% at an Intersection over Union (IoU) threshold of 0.5, a 6.3 percentage point improvement over the baseline YOLOv10. The system uses only 2.67 million parameters, demonstrating efficiency. It excels in identifying complex defects like ’rolled in scale’ and ’inclusion’. Compared to SSD and Fast R-CNN, YOLOv10n-SFDC outperforms these models in accuracy while maintaining a lightweight architecture. This system excels in automated inspection for industrial environments, offering rapid, precise defect detection. YOLOv10n-SFDC emerges as a reliable solution for the continuous monitoring and quality assurance of steel surfaces, improving the reliability and efficiency of steel manufacturing processes.

14 November 2022 in Computational Intelligence and Neuroscience
Verified
Journal Article
Open access
PDF file
Ata Jahangir MoshayediAta Jahangir Moshayedi
Ata Jahangir Moshayedi
JLJinsong Li
Jinsong Li
NSNima Sina
Nima Sina
XCXi Chen
Xi Chen
LLLiefa Liao
Liefa Liao
XXXiaoyun Xie
Xiaoyun Xie
Automated guided vehicles (AGVs) are popular subsets of robots that come in various shapes and sizes. The group’s use in the industry ranges from applications for carrying pallets, carts, and utensils to helping the elderly or transporting medicine to hospitals. Even recently, they have been used in libraries to carry books on shelves. The main part of an AGV includes its body, motor, driver, processor, and sensors, which are more or less the same in all types of AGVs, and addons vary depending on the application and the work environment. The part that affects AGV performance is the control strategy, to which researchers have shown different approaches. Using various techniques and simulations to obtain a model is the key and can help to improve and evaluate the performance of the strategy of the robot. In this study, based on the actual design of the AGV system, all data and components are described, and the simulation is performed in MATLAB software. Then, for controlling the platform based on the PID controller tuning, four methods of Ziegler Nichols, empirical, Particle Swarm Optimization (PSO), and Beetle Antennae Searching (BAS) (optimizer) are discussed, and the results are compared in the four paths including the circle, ellipse, Spiral and 8-shaped paths by observing and testing the tuned PID parameters. Finally, a series of subsequent experiences were carried out in CoppeliaSim (VREP) as a famous robot simulator to overcome the environmental constraints for the same paths that were used in Matlab based on the extracted PID values. Based on the results, the empirical methods, PSO, and BAS errors are very close together. But in general, the BAS algorithm is the fastest, and the PSO had better performance. In general, the maximum error is linked to the path of 8 shapes and the minimum is related to circle shape one. Finally, the analysis of results in different paths in both simulators shows the same results. Therefore, concerning the limited test on the real platform and using the PID coefficients obtained from the simulation shows the model’s ability for the researchers in robotic research.

1 January 2020 in Filomat
Verified
Journal Article
Open access
PDF file
Ata Jahangir MoshayediAta Jahangir Moshayedi
Ata Jahangir Moshayedi
EKEnsieh Kazemi
Ensieh Kazemi
MTMohammad Tabatabaei
Mohammad Tabatabaei
LLLiefa Liao
Liefa Liao
The main aim of this research is to propose a mathematical equation in order to reduce the model parameters based on temperature, humidity and gas density variation in metal-oxide semi-conductive sensors. Also the Arduino based Designed E-nose with the capability to change the temperature and humidity is used to obtain the real sensor?s response in various conditions. The sampling procedure consists of three sectors: fixed temperature and fixed humidity, variable temperature and fixed humidity, fixed temperature and variable humidity, which are stored in Excel software and analyzed with MATLAB. The output response is based on combination of First-Order Plus Dead Time (FOPDT) which has the Minimum Parameters system (MPS) to investigate the behavior of the sensors. Finally, after evaluating the models with the real sensor response and bi-sentence exponentials, it is suggested that the MPS model introduces fewer and simpler parameters, which helps to simulate the sensor?s behavior more accurately and consequently in order to draw a better short response.

17 November 2023
Verified
Conference Paper
Open access
TYTingjie Yu
Tingjie Yu
LLLiefa Liao
Liefa Liao
YYYiguo Yang
Yiguo Yang
WXWeihuan Xia
Weihuan Xia
ZZZhenyuan Zou
Zhenyuan Zou
In the realm of Chinese Named Entity Recognition tasks, conventional models have frequently fallen short in adequately addressing linguistic features and recognizing the essential role of context. To address this challenge, our research presents a unique Chinese Named Entity Recognition model, referred to as the LERT-DBMAT-CRF model. Initially, the LERT pre-trained language model incorporates language-informed pretraining strategies to enrich the semantic attributes in textual data. Subsequently, we apply the DBMAT module, unifying bidirectional Long Short-Term Memory networks with residual dilated convolutional networks, coordinated through a multi-head additive attention mechanism. This approach enhances feature extraction by employing the Exponential Linear Unit function, thus enhancing the model's capacity to capture temporal and spatial information relevant to semantic features. Lastly, a Conditional Random Field layer is introduced to exploit contextual information for label prediction. Experimental results demonstrate the exceptional model performance, achieving impressive F1 scores of 97.38% and 96.55% on the Resume dataset and the MSRA dataset, respectively, surpassing the performance of current mainstream models.

10 November 2025 in Buildings
Verified
Journal Article
Open access
PDF file
SLShimei Li
Shimei Li
WSWeining Song
Weining Song
TLTan Li
Tan Li
NCNanjiang Chen
Nanjiang Chen
LLLiefa Liao
Liefa Liao
RYRunmin Yin
Runmin Yin
To address information isolation and incomplete parameter extraction among multi-modal data (e.g., drawings, text, and tables) in the operation and maintenance stage of buildings, this paper proposes a multi-modal data parsing, automatic parameter extraction, and standardized integration method oriented toward 3D modeling. First, by employing vector element parsing and layer semantic analysis, the method enables structured extraction of key component geometry from architectural drawings and improves modeling accuracy via spatial topological relationship analysis. Second, by combining regular expressions, a domain-specific terminology dictionary, and a BiLSTM-CRF deep learning model, the extraction accuracy of unstructured parameters from architectural texts is significantly improved. Third, a multi-scale sliding window and geometric feature analysis are used to achieve automatic detection and parameter extraction from complex nested tables. Regarding the experimental setup: the drawings consist of a large-scale collection of DXF files stratified and randomly split into train/val/test with an approximate 8:1:1 ratio; the text set includes 1550 PDF-derived specification fragments (8:1:1 split); and the tables cover typical door/window, structural, and electrical schedules (also split ~8:1:1). F1 scores use micro-F1 (instance-level aggregation), and 95% confidence intervals and their computation are described in the main text. Experimental results show that the F1 scores for wall line, wall, and column recognition reach 98.1%, 84.9%, and 92.2%, respectively, while the F1 scores for door and window recognition are 74.3% and 76.2%. For text parameter extraction, the proposed PENet model achieves a precision of 83.56% and a recall of 86.91%. For the table task, the parameter extraction recalls for doors/windows and structure are 95.0% and 96.7%, respectively. The proposed method enables efficient parameter extraction and standardization from multi-modal architectural data, demonstrates significant advantages in handling heterogeneous data and improving modeling efficiency, and provides practical technical support for the digital reconstruction and intelligent management of existing buildings.

8 October 2026 in Scientific Reports
Verified
Journal Article
Open access
PDF file
QFQi Fan
Qi Fan
SWShouluan Wu
Shouluan Wu
LLLiefa Liao
Liefa Liao
QLQiuming Liu
Qiuming Liu
FJFU Jianglong
FU Jianglong
Object detection is a significant research direction within the field of computer vision. In recent years, Transformer-based object detection methods have achieved a commendable balance between detection accuracy and inference efficiency, with RT-DETRv2 demonstrating robust real-time detection capabilities. However, during the decoder query initialization stage, RT-DETRv2 primarily employs a global Top-K selection strategy based on encoder token classification scores. This approach tends to overlook local key regions, particularly in complex backgrounds and scenarios involving small objects, which complicates the effective retention of weak-response targets. To address this limitation, this paper proposes an enhanced method termed ROI-DETR, which is founded on ROI-constrained query selection. This method integrates the ROI Mask pass-through mechanism within the RT-DETRv2 framework and combines it with an adaptive token mask mechanism, a multi-scale budget allocation strategy, and an ROI-guided Mixed Top-K query selection strategy. These enhancements optimize the query generation process from the encoder to the decoder, thereby bolstering the model’s ability to concentrate on potential target regions. Experimental results indicate that on the COCO dataset, the Average Precision (AP) of ROI-DETR reaches 0.501, reflecting an improvement of 0.006 over RT-DETRv2, while the AP for small objects reaches 0.321, an enhancement of 0.015. Furthermore, three repeated experiments on DOTA and TinyPerson provide evidence of cross-dataset applicability. On DOTA, ROI-DETR achieves an AP of 0.259 ± 0.001 compared with 0.251 ± 0.001 for RT-DETRv2; on TinyPerson, the mean AP increases from 0.193 ± 0.002 to 0.197 ± 0.001, while the APs change remains marginal. In the ablation experiments, this paper substantiates the contribution of each sub-module to detection performance and analyzes the operational mechanisms of each module. The statistics and visualization results concerning Top-K token and ground truth (GT) matching further demonstrate that this method can enhance the coverage of target-related queries while mitigating background token interference. In summary, ROI-DETR significantly improves the selection quality of small target-related queries in complex backgrounds, offering a viable approach to optimizing the small target detection performance of real-time Transformer detectors.