Dunyun He
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
Research Subjects
Dunyun He has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
- Gui Gao
Southwest Jiaotong University
4 shared publications - Yucong He
Faculty of Geosciences and Engineering, Southwest Jiaotong University, Chengdu, China
4 shared publications - Gang Yang
Faculty of Geosciences and Engineering, Southwest Jiaotong University, Chengdu, China
3 shared publications - Tianwen Zhang
Southwest Jiaotong University
2 shared publications - Gaosheng Li
Hunan University
1 shared publication - Haoxin Zheng
Faculty of Geosciences and Engineering, Southwest Jiaotong University, Chengdu, China
1 shared publication
Latest Publications
- 15 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 23 (1545598X) , 3501905
BiFuseSR: Bifrequency Fusion With Response-Aligned Calibration and Structural Reinforcement for GNSS-S Weak-Echo Detection
Weak ship-echo detection in the Global Navigation Satellite System Scattering (GNSS-S) range–time intensity (RTI) domain is often degraded by cross-band inconsistency, time-varying response mismatch, and background inflation caused by naive dual-frequency stacking. To address these issues, a dual-frequency cooperative enhancement framework, termed BiFuseSR, is proposed for weak-echo monitoring in airborne BeiDou B1/B3 observations. First, GridSync is used to align the two bands to a common time–delay support and to normalize their background statistics, so that scale-consistent fusion can be established. Second, Doppler transfer calibration (DTC)-Align is introduced to compensate for cross-frequency Doppler/phase mismatch and improve structural consistency between bands. Third, SafeBoost is adopted to inject positive gain mainly within the target track gate while suppressing out-of-gate background accumulation. Experimental results on airborne BeiDou B1/B3 data show that, relative to the B3 baseline, the median signal-to-clutter ratio (SCR) is increased from 16.64 to 22.08 (+1.23 dB), while the 10th-percentile SCR is increased from 3.37 to 5.63, indicating improved robustness in weak and low-SNR intervals. Under controlled low-false-alarm conditions, the detection probability proxy is increased from 0.79 to 0.88 at Pfa (proxy) = 1% and from 0.72 to 0.82 at Pfa (proxy) = 0.1%, while FAproxy is maintained at about 0.01. Ablation results further show that the net gain is reduced to 0.66 dB without DTC and to 0.60 dB without SafeBoost, with the win rate dropping to 0.37 in the latter case. These results demonstrate that BiFuseSR enables reproducible dual-frequency weak-echo enhancement under controlled background statistics in GNSS-S.
- 15 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 23 (1545598X) , 3501905
BiFuseSR: Bifrequency Fusion With Response-Aligned Calibration and Structural Reinforcement for GNSS-S Weak-Echo Detection
Weak ship-echo detection in the Global Navigation Satellite System Scattering (GNSS-S) range–time intensity (RTI) domain is often degraded by cross-band inconsistency, time-varying response mismatch, and background inflation caused by naive dual-frequency stacking. To address these issues, a dual-frequency cooperative enhancement framework, termed BiFuseSR, is proposed for weak-echo monitoring in airborne BeiDou B1/B3 observations. First, GridSync is used to align the two bands to a common time–delay support and to normalize their background statistics, so that scale-consistent fusion can be established. Second, Doppler transfer calibration (DTC)-Align is introduced to compensate for cross-frequency Doppler/phase mismatch and improve structural consistency between bands. Third, SafeBoost is adopted to inject positive gain mainly within the target track gate while suppressing out-of-gate background accumulation. Experimental results on airborne BeiDou B1/B3 data show that, relative to the B3 baseline, the median signal-to-clutter ratio (SCR) is increased from 16.64 to 22.08 (+1.23 dB), while the 10th-percentile SCR is increased from 3.37 to 5.63, indicating improved robustness in weak and low-SNR intervals. Under controlled low-false-alarm conditions, the detection probability proxy is increased from 0.79 to 0.88 at Pfa (proxy) = 1% and from 0.72 to 0.82 at Pfa (proxy) = 0.1%, while FAproxy is maintained at about 0.01. Ablation results further show that the net gain is reduced to 0.66 dB without DTC and to 0.60 dB without SafeBoost, with the win rate dropping to 0.37 in the latter case. These results demonstrate that BiFuseSR enables reproducible dual-frequency weak-echo enhancement under controlled background statistics in GNSS-S.
- 30 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-14
From Fuzzy Global to Clear Local: A Focus and Super-Resolution-Guided Tiny Target Detection Method for Full-Scene Images
Detecting tiny targets, a fundamental satellite observation task, always suffers from information loss, additional computation costs, and imbalance samples in full-scene imagery. However, generic methods rely on complex deep neural networks to traverse every region of the image, irrespective of the non-uniform target distribution, severely hindering the detecting accuracy and leading to a heavy computation burden. This paper proposes a feature focusing super-resolution small target detection network (FFSSTD-Net), alleviating the above three problems. First, the proposed convolution focus detection module (CFD) uses a lightweight full-convolutional network query mechanism to adaptively search the regions with targets, eliminating unnecessary computation costs and noise from the background. Then, the proposed feature super-resolution (FSR) module guides the backbone network enhancing the representation of small target objects in high-dimensional features through an auxiliary branch to achieve a balanced optimization, alleviating the information loss caused by sub-sampling operations. Experimental results demonstrate that the FFSSTD-Net outperforms state-of-the-art generic methods, improving mAP and Fps by at least 3.14% and 10%, respectively. Meanwhile, ablation experiments indicate that the CFD and FSR modules enhance detection accuracy while accelerating full-scene image processing.
- 30 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-14
From Fuzzy Global to Clear Local: A Focus and Super-Resolution-Guided Tiny Target Detection Method for Full-Scene Images
Detecting tiny targets, a fundamental satellite observation task, always suffers from information loss, additional computation costs, and imbalance samples in full-scene imagery. However, generic methods rely on complex deep neural networks to traverse every region of the image, irrespective of the non-uniform target distribution, severely hindering the detecting accuracy and leading to a heavy computation burden. This paper proposes a feature focusing super-resolution small target detection network (FFSSTD-Net), alleviating the above three problems. First, the proposed convolution focus detection module (CFD) uses a lightweight full-convolutional network query mechanism to adaptively search the regions with targets, eliminating unnecessary computation costs and noise from the background. Then, the proposed feature super-resolution (FSR) module guides the backbone network enhancing the representation of small target objects in high-dimensional features through an auxiliary branch to achieve a balanced optimization, alleviating the information loss caused by sub-sampling operations. Experimental results demonstrate that the FFSSTD-Net outperforms state-of-the-art generic methods, improving mAP and Fps by at least 3.14% and 10%, respectively. Meanwhile, ablation experiments indicate that the CFD and FSR modules enhance detection accuracy while accelerating full-scene image processing.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 19, 1-13
GNSS-S Dual-Polarization Co-Referenced Alignment and Coherent Fusion Enhancement
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 19, 1-13
Open AccessGNSS-S Dual-Polarization Co-Referenced Alignment and Coherent Fusion Enhancement
- 17 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-16
Scattering-Attention Semi-Supervised-Guided Reinforced Region Proposal Network for Zero-Shot Detection on SAR Images
Zero-Shot Detection (ZSD), which extends conventional detection models to recognize objects from previously unseen categories, represents one of the most promising strategies for reducing data annotation costs while enhancing model generalization capabilities. Generic ZSD methods mainly focus on excavating strict mapping-transfer strategy, ignoring dataset deficiency and the heterogeneity between the visual and semantic domains in SAR imagery. This results in increased scattering noise interference and weak responses from proposal networks when dealing with unseen categories. In this paper, we proposed a scattering-attention semi-supervised-guided reinforced region proposal (SSRP) network, a detection framework that first brings the zero-shot mechanism into the realm of SAR imagery. First, to address the issue of weak cross-domain feature response, the proposal optimization module (POM) is constructed to improve the ability to capture visual information and the discrimination of the unseen classes via a teacher-student model guided by soft labels. Second, to solve the problem of scattering noise interference, the scatter excitation learning module (SELM) is proposed to reduce background noise interference and focus on the local scattering center features of the foreground examples during training to achieve a tighter cross-domain feature mapping. Then, to overcome the challenge of dataset deficiency, the picture-to-text method P2T is proposed to automatically generate descriptive terms for key attributes of SAR images, alleviating the issue of limited data availability, which is considered the main bottleneck in SAR-ZSD. We also improved the existing dataset to verify the effectiveness of the proposed method and push the development of SAR-ZSD further forward. Experimental results demonstrate that the SSRP outperforms the state-of-the-art generic methods and can improve 60% in recall and 19.59% in mAP.
- 17 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-16
Scattering-Attention Semi-Supervised-Guided Reinforced Region Proposal Network for Zero-Shot Detection on SAR Images
Zero-Shot Detection (ZSD), which extends conventional detection models to recognize objects from previously unseen categories, represents one of the most promising strategies for reducing data annotation costs while enhancing model generalization capabilities. Generic ZSD methods mainly focus on excavating strict mapping-transfer strategy, ignoring dataset deficiency and the heterogeneity between the visual and semantic domains in SAR imagery. This results in increased scattering noise interference and weak responses from proposal networks when dealing with unseen categories. In this paper, we proposed a scattering-attention semi-supervised-guided reinforced region proposal (SSRP) network, a detection framework that first brings the zero-shot mechanism into the realm of SAR imagery. First, to address the issue of weak cross-domain feature response, the proposal optimization module (POM) is constructed to improve the ability to capture visual information and the discrimination of the unseen classes via a teacher-student model guided by soft labels. Second, to solve the problem of scattering noise interference, the scatter excitation learning module (SELM) is proposed to reduce background noise interference and focus on the local scattering center features of the foreground examples during training to achieve a tighter cross-domain feature mapping. Then, to overcome the challenge of dataset deficiency, the picture-to-text method P2T is proposed to automatically generate descriptive terms for key attributes of SAR images, alleviating the issue of limited data availability, which is considered the main bottleneck in SAR-ZSD. We also improved the existing dataset to verify the effectiveness of the proposed method and push the development of SAR-ZSD further forward. Experimental results demonstrate that the SSRP outperforms the state-of-the-art generic methods and can improve 60% in recall and 19.59% in mAP.
- 12 April 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 20 (1545598X) , 1-5
Occluded Target Recognition in SAR Imagery With Scattering Excitation Learning and Channel Dropout
Deep neural networks are widely used in SAR image classification and recognition, achieving state-of-the-art performance. But it remains a challenging task to recognize occluded targets. In this letter, we propose a novel robust SAR recognition method against occlusion. Specifically, we design a scattering excitation learning module that encourages the network to learn more robust features responding to the scattering centers of targets. In addition, we adopt a random feature channel dropout technique which can further improve robustness to occlusion. Our method makes the network more robust against occlusion but without any occlusion-simulated data for training. Experimental results on MSTAR dataset shows that our proposed method achieves remarkably improved robustness even under severe occlusions. Code is made available at https://github.com/koervcor/SEL-CD.
- 12 April 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 20 (1545598X) , 1-5
Occluded Target Recognition in SAR Imagery With Scattering Excitation Learning and Channel Dropout
Deep neural networks are widely used in SAR image classification and recognition, achieving state-of-the-art performance. But it remains a challenging task to recognize occluded targets. In this letter, we propose a novel robust SAR recognition method against occlusion. Specifically, we design a scattering excitation learning module that encourages the network to learn more robust features responding to the scattering centers of targets. In addition, we adopt a random feature channel dropout technique which can further improve robustness to occlusion. Our method makes the network more robust against occlusion but without any occlusion-simulated data for training. Experimental results on MSTAR dataset shows that our proposed method achieves remarkably improved robustness even under severe occlusions. Code is made available at https://github.com/koervcor/SEL-CD.