Jun Ying
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jun Ying has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
- Junlei Song
School of Artificial Intelligence and Automation, China University of Geosciences, Wuhan 430074, China
2 shared publications - Shikun Yin
Hubei Provincial Center for Flood and Drought Disaster Prevention, Wuhan 430074, China
2 shared publications - Yi Zhou
Hubei Provincial Center for Flood and Drought Disaster Prevention, Wuhan 430074, China
2 shared publications - Bei Zhou
School of Artificial Intelligence and Automation, China University of Geosciences, Wuhan 430074, China
1 shared publication - Chenggang Zhou
Faculty of Materials Science and Chemistry, China University of Geosciences, Wuhan 430074, China
1 shared publication - Fang Jin
School of Automation, China University of Geosciences 1 , Wuhan 430074
1 shared publication
Latest Publications
- Open Access
- 18 September 2026
- Published by MDPI AG in Electronics
- Vol. 15 (18) , 4271
Enhanced Lightweight Image Super-Resolution via Residual Aggregation and Wavelet Loss
Image super-resolution (SR), which aims to reconstruct a high-resolution image from a low-resolution input, has progressed from convolutional neural networks (CNNs) to transformer-based architectures. Despite this progress, lightweight transformer SR remains challenging: local or window-based operations provide limited long-range interaction, conventional query-key-value projections introduce parameter and computational redundancy, and pixel-domain loss alone provides insufficient frequency-domain constraints on fine structures. This study presents RAW, a lightweight SR network based on residual aggregation and wavelet loss. RAW uses local aggregation to preserve neighborhood textures, mesoscale grouped-residual attention to reduce projection redundancy while modeling regional dependencies, and non-local sparse aggregation to capture long-range information at a controlled cost. By integrating stationary-wavelet-transform loss with RGB-domain L1 loss, the model supervises structural and high-frequency information without adding an inference branch. Experiments on standard benchmarks demonstrate a competitive trade-off between reconstruction quality and computational complexity. For 4× SR, RAW reduces the numbers of parameters and MACs by 14.3% and 15.4%, respectively, relative to the baseline, while improving the PSNR and SSIM on Manga109 by 0.22 dB and 0.0015, respectively.
- 18 September 2026
- Published by MDPI AG in Electronics
- Vol. 15 (18) , 4271
Open AccessEnhanced Lightweight Image Super-Resolution via Residual Aggregation and Wavelet Loss
Image super-resolution (SR), which aims to reconstruct a high-resolution image from a low-resolution input, has progressed from convolutional neural networks (CNNs) to transformer-based architectures. Despite this progress, lightweight transformer SR remains challenging: local or window-based operations provide limited long-range interaction, conventional query-key-value projections introduce parameter and computational redundancy, and pixel-domain loss alone provides insufficient frequency-domain constraints on fine structures. This study presents RAW, a lightweight SR network based on residual aggregation and wavelet loss. RAW uses local aggregation to preserve neighborhood textures, mesoscale grouped-residual attention to reduce projection redundancy while modeling regional dependencies, and non-local sparse aggregation to capture long-range information at a controlled cost. By integrating stationary-wavelet-transform loss with RGB-domain L1 loss, the model supervises structural and high-frequency information without adding an inference branch. Experiments on standard benchmarks demonstrate a competitive trade-off between reconstruction quality and computational complexity. For 4× SR, RAW reduces the numbers of parameters and MACs by 14.3% and 15.4%, respectively, relative to the baseline, while improving the PSNR and SSIM on Manga109 by 0.22 dB and 0.0015, respectively.
- Open Access
Ultrasonic Point Cloud-Based Segmentation and Quantification of Internal Defects in Concrete Structures
Accurate detection of internal defects in concrete structures is critical to structural safety. This study proposes a three-stage method for segmentation and geometric quantification of defects from 3D ultrasonic point clouds to support non-destructive testing (NDT). Sparse point clouds are densified by grid-based, acoustically constrained trilinear interpolation in the linear amplitude domain to preserve defect-boundary fidelity. An Amplitude-aware PointNet (AAPNet) integrates spatial coordinates with normalized and thresholded echo amplitude for defect segmentation. Connected-component analysis on voxelized defect regions enables estimation of length, width, height, and volume. Laboratory specimens containing voids, cracks, and delaminations were scanned with an Elop Insight 3D ultrasonic system. Segmentation was evaluated by nine-fold leave-one-specimen-out cross-validation (LOSO-CV) on 250 scans from nine independent single-defect specimens, yielding a mean IoU of 82.75 ± 2.66% (precision 92.03 ± 1.58%, recall 89.11 ± 1.73%, F1 90.54 ± 1.59%). Dimensional quantification on embedded defects gave relative volume errors of 2.6–14.9% and axis-aligned deviations within 2.0 cm.
Open AccessUltrasonic Point Cloud-Based Segmentation and Quantification of Internal Defects in Concrete Structures
Accurate detection of internal defects in concrete structures is critical to structural safety. This study proposes a three-stage method for segmentation and geometric quantification of defects from 3D ultrasonic point clouds to support non-destructive testing (NDT). Sparse point clouds are densified by grid-based, acoustically constrained trilinear interpolation in the linear amplitude domain to preserve defect-boundary fidelity. An Amplitude-aware PointNet (AAPNet) integrates spatial coordinates with normalized and thresholded echo amplitude for defect segmentation. Connected-component analysis on voxelized defect regions enables estimation of length, width, height, and volume. Laboratory specimens containing voids, cracks, and delaminations were scanned with an Elop Insight 3D ultrasonic system. Segmentation was evaluated by nine-fold leave-one-specimen-out cross-validation (LOSO-CV) on 250 scans from nine independent single-defect specimens, yielding a mean IoU of 82.75 ± 2.66% (precision 92.03 ± 1.58%, recall 89.11 ± 1.73%, F1 90.54 ± 1.59%). Dimensional quantification on embedded defects gave relative volume errors of 2.6–14.9% and axis-aligned deviations within 2.0 cm.