Mengyu Sun
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Mengyu Sun has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
Chu Wang
School of Electronics and Information Engineering, Xi’an Technological University, Xi’an, China
1 shared publicationLiangliang Li
School of Institute for Interdisciplinary and Innovation Research, Xi’an Technological University, Xi’an, China
1 shared publication- Liangliang Li
Institute for Interdisciplinary and Innovation Research, Xi’an Technological University, Xi’an 710021, China
1 shared publication - Peng Wang
Institute for Interdisciplinary and Innovation Research, Xi'an Technological University, Xi'an 710021, China
1 shared publication - Peng Wang
Institute for Interdisciplinary and Innovation Research, Xian Technological University, Xian, China
1 shared publication - Peng Wang
School of Institute for Interdisciplinary and Innovation Research, Xi’an Technological University, Xi’an, China
1 shared publication
Latest Publications
- Open Access
- 22 May 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-18
High resolution weld semantic defect detection algorithm based on integrated double U structure
As a key industrial equipment, the welding quality of pressure vessels is directly related to the operation safety, and X-ray nondestructive testing is an important means to evaluate the weld quality. In this study, a high-resolution semantic segmentation algorithm based on a double U-shaped network was proposed to meet the requirements of pressure vessel weld defect detection, and accurate detection was realized through innovative three-stage processing flow. Firstly, the weld area was automatically extracted from the original X-ray films (3000+×800 + pixels) by the Gaussian positioning algorithm. Then, a multi-image hybrid stitching technology was proposed to reconstruct the long weld into a standard size of 1500 × 1500, which not only expanded the WSCR data set, but also effectively improved the data imbalance problem. In terms of model architecture, by improving U2Netp and UNet networks, MC-SPP module (multi-connection spatial pyramid pooling), RMAG module (residual multi-add gating recurrent unit), HDC-CBAM module (hybrid dilated convolution-convolutional block attention) and CCM module (cross-layer connection fusion) were integrated to form a cascade network with multi-level feature fusion capability. The experimental results showed that the model could effectively segment the defects such as cracks, pores, slag inclusion, incomplete fusion and incomplete penetration in the weld film, and the mIOU value reached 78.1, which was significantly superior to the existing methods, and provided an efficient and reliable solution for welding defect detection of pressure vessels, which had important engineering application value.
- 22 May 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-18
Open AccessHigh resolution weld semantic defect detection algorithm based on integrated double U structure
As a key industrial equipment, the welding quality of pressure vessels is directly related to the operation safety, and X-ray nondestructive testing is an important means to evaluate the weld quality. In this study, a high-resolution semantic segmentation algorithm based on a double U-shaped network was proposed to meet the requirements of pressure vessel weld defect detection, and accurate detection was realized through innovative three-stage processing flow. Firstly, the weld area was automatically extracted from the original X-ray films (3000+×800 + pixels) by the Gaussian positioning algorithm. Then, a multi-image hybrid stitching technology was proposed to reconstruct the long weld into a standard size of 1500 × 1500, which not only expanded the WSCR data set, but also effectively improved the data imbalance problem. In terms of model architecture, by improving U2Netp and UNet networks, MC-SPP module (multi-connection spatial pyramid pooling), RMAG module (residual multi-add gating recurrent unit), HDC-CBAM module (hybrid dilated convolution-convolutional block attention) and CCM module (cross-layer connection fusion) were integrated to form a cascade network with multi-level feature fusion capability. The experimental results showed that the model could effectively segment the defects such as cracks, pores, slag inclusion, incomplete fusion and incomplete penetration in the weld film, and the mIOU value reached 78.1, which was significantly superior to the existing methods, and provided an efficient and reliable solution for welding defect detection of pressure vessels, which had important engineering application value.
- Open Access
MonoDFNet: Monocular 3D Object Detection with Depth Fusion and Adaptive Optimization
Monocular 3D object detection refers to detecting 3D objects using a single camera. This approach offers low sensor costs, high resolution, and rich texture information, making it widely adopted. However, monocular sensors face challenges from environmental factors like occlusion and truncation, leading to reduced detection accuracy. Additionally, the lack of depth information poses significant challenges for predicting 3D positions. To address these issues, this paper presents a monocular 3D object detection method based on improvements to MonoCD, designed to enhance detection accuracy and robustness in complex environments. In order to effectively obtain and integrate depth information, this paper designs a multi-branch depth prediction with weight sharing module. Furthermore, an adaptive focus mechanism is proposed to emphasize target regions while minimizing interference from irrelevant areas. The experimental results demonstrate that MonoDFNet achieves significant improvements over existing methods, with AP3D gains of +4.09% (Easy), +2.78% (Moderate), and +1.63% (Hard), confirming its effectiveness in 3D object detection.
Open AccessMonoDFNet: Monocular 3D Object Detection with Depth Fusion and Adaptive Optimization
Monocular 3D object detection refers to detecting 3D objects using a single camera. This approach offers low sensor costs, high resolution, and rich texture information, making it widely adopted. However, monocular sensors face challenges from environmental factors like occlusion and truncation, leading to reduced detection accuracy. Additionally, the lack of depth information poses significant challenges for predicting 3D positions. To address these issues, this paper presents a monocular 3D object detection method based on improvements to MonoCD, designed to enhance detection accuracy and robustness in complex environments. In order to effectively obtain and integrate depth information, this paper designs a multi-branch depth prediction with weight sharing module. Furthermore, an adaptive focus mechanism is proposed to emphasize target regions while minimizing interference from irrelevant areas. The experimental results demonstrate that MonoDFNet achieves significant improvements over existing methods, with AP3D gains of +4.09% (Easy), +2.78% (Moderate), and +1.63% (Hard), confirming its effectiveness in 3D object detection.