Dinghai Wu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Dinghai Wu has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Chun Lv
Department of Vehicle and Electrical Engineering, Army Engineering University, Shijiazhuang 050003, China
1 shared publication - Guichao Liu
Shijiazhuang Campus of the Army Engineering University of PLA, Shijiazhuang 050051, China
1 shared publication Guoquan Ren
Army Engineering University of PLA, Shijiazhuang Campus, Shijiazhuang, China
1 shared publication- Hongbo Fan
Department of Vehicle and Electrical Engineering, Army Engineering University of PLA, Shijiazhuang, China
1 shared publication - Huaiguang Wang
Army Engineering University of PLA, Shijiazhuang Campus, Shijiazhuang, China
1 shared publication Jun Li
Army Engineering University of PLA, Shijiazhuang Campus, Shijiazhuang, China
1 shared publication
Latest Publications
- Open Access
- 10 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 91132-91144
A 2 -Detector: An All-Weather Robust Armored Target Detector Based on an Enhance-and-Purify Framework
Adverse weather conditions, such as rain, fog, and low illumination, can severely degrade the visual features required for armored target detection, leading to significant performance drops in real-world deployments. We propose A²-Detector, a physics-prior-inspired framework for feature enhancement and pu-rification. The framework consists of two complementary components: HMRFRM progressively enhances degraded representations through multi-scale feature refinement, while GCPP suppresses noise propagation during multi-scale fusion. In addition, Shape-IoU loss is adopted to improve localization stability under blurred boundaries and low contrast. Extensive experiments on Front-view armored target dataset (FATD) show that A²-Detector achieves 93.5% AP50 with only 8.9% performance degradation under synthetic weather degradations. Cross-domain evaluation on BDD-100K further verifies the generalizability of the proposed design, achieving 83.6% AP50 for civilian-vehicle detection. These results indicate that the pro-posed framework improves detection robustness by learning task-oriented feature enhancement and purifica-tion under physics-informed degradation priorities.
- 10 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 91132-91144
Open AccessA 2 -Detector: An All-Weather Robust Armored Target Detector Based on an Enhance-and-Purify Framework
Adverse weather conditions, such as rain, fog, and low illumination, can severely degrade the visual features required for armored target detection, leading to significant performance drops in real-world deployments. We propose A²-Detector, a physics-prior-inspired framework for feature enhancement and pu-rification. The framework consists of two complementary components: HMRFRM progressively enhances degraded representations through multi-scale feature refinement, while GCPP suppresses noise propagation during multi-scale fusion. In addition, Shape-IoU loss is adopted to improve localization stability under blurred boundaries and low contrast. Extensive experiments on Front-view armored target dataset (FATD) show that A²-Detector achieves 93.5% AP50 with only 8.9% performance degradation under synthetic weather degradations. Cross-domain evaluation on BDD-100K further verifies the generalizability of the proposed design, achieving 83.6% AP50 for civilian-vehicle detection. These results indicate that the pro-posed framework improves detection robustness by learning task-oriented feature enhancement and purifica-tion under physics-informed degradation priorities.
- 23 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 877-882
Mechanical Fault Diagnosis based on Dual-tree Complex Wavelet Packet Time-frequency Distribution and Residual Network Transfer Learning
Aiming at the problem of large amount of calculation for the time-frequency distribution of mechanical vibration signals with good representation ability, a mechanical fault diagnosis method based on dual-tree complex wavelet packet time-frequency distribution and residual network transfer learning is proposed. Firstly, the high-frequency and low-frequency components of the signal are decomposed based on the dual-tree complex wavelet packet with translation invariance, and the discrete time-frequency distribution of the mechanical vibration signal is efficiently obtained. In view of the fact that ResNet uses residual modules and cross-layer connections to build a deep network structure, it has strong image recognition and classification capabilities, Residual network transfer learning is used to train and classify the obtained time-frequency distribution images, and the Western Reserve University bearing dataset and the engines measured signal are used to verify the method proposed in this paper, and it can obtain ideal fault diagnosis results under different input images sizes and a small number of training sample conditions.
- 23 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 877-882
Mechanical Fault Diagnosis based on Dual-tree Complex Wavelet Packet Time-frequency Distribution and Residual Network Transfer Learning
Aiming at the problem of large amount of calculation for the time-frequency distribution of mechanical vibration signals with good representation ability, a mechanical fault diagnosis method based on dual-tree complex wavelet packet time-frequency distribution and residual network transfer learning is proposed. Firstly, the high-frequency and low-frequency components of the signal are decomposed based on the dual-tree complex wavelet packet with translation invariance, and the discrete time-frequency distribution of the mechanical vibration signal is efficiently obtained. In view of the fact that ResNet uses residual modules and cross-layer connections to build a deep network structure, it has strong image recognition and classification capabilities, Residual network transfer learning is used to train and classify the obtained time-frequency distribution images, and the Western Reserve University bearing dataset and the engines measured signal are used to verify the method proposed in this paper, and it can obtain ideal fault diagnosis results under different input images sizes and a small number of training sample conditions.
- Open Access
- 6 March 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 47490-47499
Gear Fault Feature Extraction Based on Fuzzy Function and Improved Hu Invariant Moments
To intelligently identify gear fault types on the basis of gear vibration signals, due to the non-linear and non-stationary characteristics of gear vibration signals, a fuzzy function is used to represent the gear vibration signals in different states as two-dimensional time-frequency images. To solve the problem that the recognition effect of discontinuity area in the image by traditional Hu invariant moments is not ideal, improved Hu invariant moments are proposed, and the feature parameters of time-frequency images of gear vibration signals are extracted based on the improved Hu invariant moments. Different intelligent classifiers are used to recognize the gear vibration signals in different states. The recognition accuracy is higher by improved Hu invariant moments than by traditional Hu invariant moments, which shows that the method of gear fault feature extraction based on a fuzzy function and improved Hu invariant moments is quite ideal, and can be used in intelligent diagnosis of gear faults.
- 6 March 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 47490-47499
Open AccessGear Fault Feature Extraction Based on Fuzzy Function and Improved Hu Invariant Moments
To intelligently identify gear fault types on the basis of gear vibration signals, due to the non-linear and non-stationary characteristics of gear vibration signals, a fuzzy function is used to represent the gear vibration signals in different states as two-dimensional time-frequency images. To solve the problem that the recognition effect of discontinuity area in the image by traditional Hu invariant moments is not ideal, improved Hu invariant moments are proposed, and the feature parameters of time-frequency images of gear vibration signals are extracted based on the improved Hu invariant moments. Different intelligent classifiers are used to recognize the gear vibration signals in different states. The recognition accuracy is higher by improved Hu invariant moments than by traditional Hu invariant moments, which shows that the method of gear fault feature extraction based on a fuzzy function and improved Hu invariant moments is quite ideal, and can be used in intelligent diagnosis of gear faults.