Xinyu Zuo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Xinyu Zuo has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Ahmed D. Almutairi
Department of Civil Engineering, College of Engineering, Qassim University, Buraydah, Saudi Arabia
3 shared publications - Yiqing Dai
College of Civil Engineering, Fuzhou University, Fuzhou 350108, China
2 shared publications - Chao Yu
College of Civil Engineering, Fuzhou University, Fuzhou 350108, China
1 shared publication - Gang Wang
The Hong Kong University of Science and Technology
1 shared publication - Guangsheng Zeng
College of Civil Engineering, Fuzhou University, Fuzhou 350108, China
1 shared publication - Muneer K. Saeed
Department of Civil Engineering, College of Engineering, Qassim University, Buraydah 51452, Saudi Arabia
1 shared publication
Latest Publications
- Open Access
- 9 October 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1) , 33214
Enhanced YOLOv8s for lightweight asphalt pavement crack detection under complex road conditions
Cracking is a common pavement distress and has a significant impact on structural integrity, traffic safety, and maintenance strategies throughout the entire pavement life cycle. This study presents an improved YOLOv8s-based framework that can effectively and accurately detect asphalt pavement cracks. The proposed model is designed to overcome the main limitations of current methods, such as limited performance, low detection accuracy and limited ability to detect small-scale cracks under complex pavement conditions. To optimize the bounding-box regression and improve the bounding-box localization accuracy of crack instances, a Wise-IoU (WIoU) loss function is introduced. Moreover, Ghost Shuffle Convolution (GSConv) is incorporated into the backbone to improve feature extraction and decrease computational complexity. Furthermore, an improved C2f module is designed to improve the detection of small-scale and low-contrast crack instances. The effectiveness of the proposed method is demonstrated through experimental evaluations on an asphalt pavement crack dataset constructed by combining field-collected images with publicly available pavement crack datasets. Conventional data augmentation operations were further applied to the training images to improve data diversity and model generalization. Experimental results show that the proposed method achieves consistent improvement over the original YOLOv8s model by up to 3.2% points in mean Average Precision (mAP) and 4.0% points in F1-score. In addition, the proposed model performs better in complex pavement environments and can better extract crack-related features. The results indicate that the improved YOLOv8s framework provides an accurate, efficient and intelligent solution for asphalt pavement crack detection, and is a promising method for automated pavement distress evaluation and management.
- 9 October 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1) , 33214
Open AccessEnhanced YOLOv8s for lightweight asphalt pavement crack detection under complex road conditions
Cracking is a common pavement distress and has a significant impact on structural integrity, traffic safety, and maintenance strategies throughout the entire pavement life cycle. This study presents an improved YOLOv8s-based framework that can effectively and accurately detect asphalt pavement cracks. The proposed model is designed to overcome the main limitations of current methods, such as limited performance, low detection accuracy and limited ability to detect small-scale cracks under complex pavement conditions. To optimize the bounding-box regression and improve the bounding-box localization accuracy of crack instances, a Wise-IoU (WIoU) loss function is introduced. Moreover, Ghost Shuffle Convolution (GSConv) is incorporated into the backbone to improve feature extraction and decrease computational complexity. Furthermore, an improved C2f module is designed to improve the detection of small-scale and low-contrast crack instances. The effectiveness of the proposed method is demonstrated through experimental evaluations on an asphalt pavement crack dataset constructed by combining field-collected images with publicly available pavement crack datasets. Conventional data augmentation operations were further applied to the training images to improve data diversity and model generalization. Experimental results show that the proposed method achieves consistent improvement over the original YOLOv8s model by up to 3.2% points in mean Average Precision (mAP) and 4.0% points in F1-score. In addition, the proposed model performs better in complex pavement environments and can better extract crack-related features. The results indicate that the improved YOLOv8s framework provides an accurate, efficient and intelligent solution for asphalt pavement crack detection, and is a promising method for automated pavement distress evaluation and management.
- Open Access
Improved BP Neural Network Ensemble Model for Asphalt Pavement Performance Prediction
With the growing complexity and variability of the operational environment of asphalt pavements and the continuous increase in traffic loads, traditional pavement performance prediction models cannot accurately depict the nonlinear degradation process of pavement performance with the passage of time. Therefore, a novel approach is proposed in this paper to forecast the service performance of asphalt pavements accurately. This method optimises a Backpropagation (BP) neural network using the Levenberg–Marquardt (LM) algorithm. Seven main influencing factors were selected as the input parameters to construct the prediction model, and the performance of the prediction model was evaluated. The parameters considered in this analysis are: road age, average daily traffic volume for one year, annual temperature range, annual precipitation, relative humidity, pavement thickness and pavement surface compressive strength. Through these factors cumulatively, the model is able to predict and evaluate the road condition and Pavement Quality Index (PQI) accurately. The results indicate that the proposed model is better than the baseline model of traditional BP neural networks in predicting the Road Condition Index (RCI), with a Mean Absolute Error (MAE) of 0.395. This is an important reference to predict the service performance of asphalt pavements and validate the effectiveness of the model.
Open AccessImproved BP Neural Network Ensemble Model for Asphalt Pavement Performance Prediction
With the growing complexity and variability of the operational environment of asphalt pavements and the continuous increase in traffic loads, traditional pavement performance prediction models cannot accurately depict the nonlinear degradation process of pavement performance with the passage of time. Therefore, a novel approach is proposed in this paper to forecast the service performance of asphalt pavements accurately. This method optimises a Backpropagation (BP) neural network using the Levenberg–Marquardt (LM) algorithm. Seven main influencing factors were selected as the input parameters to construct the prediction model, and the performance of the prediction model was evaluated. The parameters considered in this analysis are: road age, average daily traffic volume for one year, annual temperature range, annual precipitation, relative humidity, pavement thickness and pavement surface compressive strength. Through these factors cumulatively, the model is able to predict and evaluate the road condition and Pavement Quality Index (PQI) accurately. The results indicate that the proposed model is better than the baseline model of traditional BP neural networks in predicting the Road Condition Index (RCI), with a Mean Absolute Error (MAE) of 0.395. This is an important reference to predict the service performance of asphalt pavements and validate the effectiveness of the model.
- Open Access
Improved Dual-Module YOLOv8 Algorithm for Building Crack Detection
Building cracks are significant indicators of structural integrity. Conventional fracture detection methodologies, however, are characterized by extended durations, significant labor requirements, and limitations in both precision and operational effectiveness. Findings are also subject to subjective and technical constraints inherent in manual assessments. To overcome these challenges, this paper introduces an enhanced YOLOv8-based methodology for developing a building crack detection system, thereby achieving high precision, operational efficiency, and cost-effectiveness. Initially, classified and segmented datasets of building fractures were obtained from field photography, online image aggregation, and open-source databases, thereby providing the basis for training the experimental model. Subsequently, the Swin Transformer window multi-head self-attention mechanism was implemented to augment small-object recognition capabilities and reduce computational demands, thereby enabling the development of an enhanced image segmentation module. Utilizing the U-Net’s segmentation capabilities, a rotated split method was implemented to quantify fracture width and derive geometric parameters from the segmented crack regions. In order to evaluate the effectiveness of the model, two experiments were conducted: one to demonstrate the performance of the classification category and the other to show the capabilities of the segmentation category. The result is that the proposed model has high accuracy and efficiency in the frac detection task. This approach effectively enhances fracture detection in structural safety evaluations of these buildings, providing technical support for relevant management decisions.
Open AccessImproved Dual-Module YOLOv8 Algorithm for Building Crack Detection
Building cracks are significant indicators of structural integrity. Conventional fracture detection methodologies, however, are characterized by extended durations, significant labor requirements, and limitations in both precision and operational effectiveness. Findings are also subject to subjective and technical constraints inherent in manual assessments. To overcome these challenges, this paper introduces an enhanced YOLOv8-based methodology for developing a building crack detection system, thereby achieving high precision, operational efficiency, and cost-effectiveness. Initially, classified and segmented datasets of building fractures were obtained from field photography, online image aggregation, and open-source databases, thereby providing the basis for training the experimental model. Subsequently, the Swin Transformer window multi-head self-attention mechanism was implemented to augment small-object recognition capabilities and reduce computational demands, thereby enabling the development of an enhanced image segmentation module. Utilizing the U-Net’s segmentation capabilities, a rotated split method was implemented to quantify fracture width and derive geometric parameters from the segmented crack regions. In order to evaluate the effectiveness of the model, two experiments were conducted: one to demonstrate the performance of the classification category and the other to show the capabilities of the segmentation category. The result is that the proposed model has high accuracy and efficiency in the frac detection task. This approach effectively enhances fracture detection in structural safety evaluations of these buildings, providing technical support for relevant management decisions.
- Open Access
Enhanced YOLOv8n-Based Three-Module Lightweight Helmet Detection System
Maintaining a safe working environment for construction workers is critical to the improvement of urban areas. Several issues plague the present safety helmet detection technologies utilized on construction sites. Some of these issues include low accuracy, expensive deployment of edge devices, and complex backgrounds. To overcome these obstacles, this paper introduces a detection method that is both efficient and based on an improved version of YOLOv8n. Three components make up the superior algorithm: the C2f-SCConv architecture, the Partial Convolutional Detector (PCD), and Coordinate Attention (CA). Detection, redundancy reduction, and feature localization accuracy are all improved with coordinate attention. To further enhance feature quality, decrease computing cost, and make corrections more effective, a Partial Convolution detector is subsequently constructed. Feature refinement and feature representation are made more effective by using C2f-SCConv instead of the bottleneck C2f module. In comparison to its predecessor, the upgraded YOLOv8n is superior in every respect. It reduced model size by 2.21 MB, increased frame rate by 12.6 percent, decreased FLOPs by 49.9 percent, and had an average accuracy of 94.4 percent. This method is more efficient, quicker, and cheaper to set up on-site than conventional helmet-detection algorithms.
Open AccessEnhanced YOLOv8n-Based Three-Module Lightweight Helmet Detection System
Maintaining a safe working environment for construction workers is critical to the improvement of urban areas. Several issues plague the present safety helmet detection technologies utilized on construction sites. Some of these issues include low accuracy, expensive deployment of edge devices, and complex backgrounds. To overcome these obstacles, this paper introduces a detection method that is both efficient and based on an improved version of YOLOv8n. Three components make up the superior algorithm: the C2f-SCConv architecture, the Partial Convolutional Detector (PCD), and Coordinate Attention (CA). Detection, redundancy reduction, and feature localization accuracy are all improved with coordinate attention. To further enhance feature quality, decrease computing cost, and make corrections more effective, a Partial Convolution detector is subsequently constructed. Feature refinement and feature representation are made more effective by using C2f-SCConv instead of the bottleneck C2f module. In comparison to its predecessor, the upgraded YOLOv8n is superior in every respect. It reduced model size by 2.21 MB, increased frame rate by 12.6 percent, decreased FLOPs by 49.9 percent, and had an average accuracy of 94.4 percent. This method is more efficient, quicker, and cheaper to set up on-site than conventional helmet-detection algorithms.