Jiayi Zhao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Jiayi Zhao has a total of 24 co-authors. Here are the top co-authors by number of shared publications:
- Guisheng Liao
Xidian University
6 shared publications - Xuepan Zhang
National Key Laboratory of Radar Signal Processing, Xidian University, Xian, China
4 shared publications - Yaxing Yue
Xidian University
4 shared publications - Zhiguo Shi
Zhejiang University
3 shared publications - Kai Fang
Zhejiang A&F University
2 shared publications - Sining Liu
National Key Laboratory of Radar Signal Processing, Xidian University, Xi'an, China
2 shared publications
Latest Publications
- Open Access
- 2 June 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 (19391404) , 20385-20400
An LLM-Enhanced Anomaly Detection Method for Identifying RFI in Multitemporal SAR Data
With the increasing complexity of the electromagnetic environment, synthetic aperture radar (SAR) sensors are frequently affected by radio frequency interference (RFI), which manifests as de-focused, block-like anomaly artifacts in SAR images. Traditional intelligent detection methods rely on extensive annotation, which is impractical due to the massive volume of data and the extreme scarcity of RFI-contaminated samples. To address these challenges, this paper proposes a large language model (LLM) enhanced unsupervised RFI detection framework, leveraging the unprecedented capabilities of LLM in multi-modal reasoning. First, to incorporate semantic guidance, we introduce a text encoder module based on bidirectional long short-term memory. This module encodes LLM-generated image quality assessment reports into high-dimensional feature vectors, providing semantic constraints for anomaly detection. Second, a fusion metric is introduced to synthesize the dual-polarization quick-look image into a single-channel image, efficiently reducing input data volume while preserving interference features. Third, a temporal filter is applied to the multi-temporal synthesized image sequence to construct clean background images as a training set. Finally, a multimodal anomaly detection network, named RFI-MADNet, is proposed. It integrates the textual embeddings into a nested autoencoder-generative adversarial network architecture with self-attention and refined loss for intelligent RFI detection and localization. The proposed method effectively achieves robust detection on class-imbalanced real-world SAR data without requiring manual labeling. Experimental results show that the proposed method achieves superior RFI detection and competitive localization performance across various regions compared to current methods, with improvements of at least 14% in the average area under the curve metric and over 1% in the mean intersection over union metric.
- 2 June 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 (19391404) , 20385-20400
Open AccessAn LLM-Enhanced Anomaly Detection Method for Identifying RFI in Multitemporal SAR Data
With the increasing complexity of the electromagnetic environment, synthetic aperture radar (SAR) sensors are frequently affected by radio frequency interference (RFI), which manifests as de-focused, block-like anomaly artifacts in SAR images. Traditional intelligent detection methods rely on extensive annotation, which is impractical due to the massive volume of data and the extreme scarcity of RFI-contaminated samples. To address these challenges, this paper proposes a large language model (LLM) enhanced unsupervised RFI detection framework, leveraging the unprecedented capabilities of LLM in multi-modal reasoning. First, to incorporate semantic guidance, we introduce a text encoder module based on bidirectional long short-term memory. This module encodes LLM-generated image quality assessment reports into high-dimensional feature vectors, providing semantic constraints for anomaly detection. Second, a fusion metric is introduced to synthesize the dual-polarization quick-look image into a single-channel image, efficiently reducing input data volume while preserving interference features. Third, a temporal filter is applied to the multi-temporal synthesized image sequence to construct clean background images as a training set. Finally, a multimodal anomaly detection network, named RFI-MADNet, is proposed. It integrates the textual embeddings into a nested autoencoder-generative adversarial network architecture with self-attention and refined loss for intelligent RFI detection and localization. The proposed method effectively achieves robust detection on class-imbalanced real-world SAR data without requiring manual labeling. Experimental results show that the proposed method achieves superior RFI detection and competitive localization performance across various regions compared to current methods, with improvements of at least 14% in the average area under the curve metric and over 1% in the mean intersection over union metric.
- 8 January 2026
- Published by Elsevier in Signal Processing
- Vol. 243
A weighted coherent integration method for weak target detection based on active-passive radar
- 8 January 2026
- Published by Elsevier in Signal Processing
- Vol. 243
A weighted coherent integration method for weak target detection based on active-passive radar
- 10 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-16
A Joint Framework of Wavelet Filtering and Fast GSVT-LRSD Algorithm for SAR Narrowband Pulsed RFI Suppression
As the electromagnetic spectrum becomes increasingly crowded in recent years, synthetic aperture radar (SAR) is confronted with an escalating amount of radio frequency interference (RFI). In civilian SAR satellite data, narrowband pulsed RFI (PRFI) is a prevalent interference type that significantly degrades the interpretability of SAR images. Among most approaches for suppressing PRFI, notch filtering methods face significant challenges in threshold selection of signal intensity. Conversely, low-rank and sparse decomposition (LRSD) algorithms, though free of threshold selection, often struggle to satisfy the required low-rank conditions. These limitations underscore the necessity of developing more robust interference suppression methods. In this paper, we propose a joint framework of wavelet filtering and fast generalized singular value thresholding based LRSD method (FGSVT-LRSD) to suppress narrowband PRFI in rangefrequency and azimuth-time domain of SAR single-look complex (SLC) data. First, a wavelet domain notch filtering (WNF) method is employed to extract the strong spectral components that are primarily composed of PRFI in the two-dimensional (2D) range spectrum of SAR SLC data, while simultaneously protecting the spectrum of low-energy useful signals. Then, the FGSVT-LRSD method is applied to the extracted strong spectral components to efficiently separate the PRFI from the useful signals. By strategically integrating these two approaches, the proposed framework simultaneously exploits the high-intensity and low-rank properties of PRFI for more precise separation, significantly reduces the sensitivity of threshold selection in the WNF process, and facilitates the low-rank conditions required by the FGSVT-LRSD method. Finally, the separated PRFI spectrum is subtracted from the original 2D range spectrum, resulting in the RFI-suppressed SAR image. Experimental results based on both simulated and measured spaceborne SAR data will be presented to demonstrate that, compared to existing methods, the proposed approach exhibits superior PRFI suppression capabilities and effectively preserves useful signals.
- 10 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-16
A Joint Framework of Wavelet Filtering and Fast GSVT-LRSD Algorithm for SAR Narrowband Pulsed RFI Suppression
As the electromagnetic spectrum becomes increasingly crowded in recent years, synthetic aperture radar (SAR) is confronted with an escalating amount of radio frequency interference (RFI). In civilian SAR satellite data, narrowband pulsed RFI (PRFI) is a prevalent interference type that significantly degrades the interpretability of SAR images. Among most approaches for suppressing PRFI, notch filtering methods face significant challenges in threshold selection of signal intensity. Conversely, low-rank and sparse decomposition (LRSD) algorithms, though free of threshold selection, often struggle to satisfy the required low-rank conditions. These limitations underscore the necessity of developing more robust interference suppression methods. In this paper, we propose a joint framework of wavelet filtering and fast generalized singular value thresholding based LRSD method (FGSVT-LRSD) to suppress narrowband PRFI in rangefrequency and azimuth-time domain of SAR single-look complex (SLC) data. First, a wavelet domain notch filtering (WNF) method is employed to extract the strong spectral components that are primarily composed of PRFI in the two-dimensional (2D) range spectrum of SAR SLC data, while simultaneously protecting the spectrum of low-energy useful signals. Then, the FGSVT-LRSD method is applied to the extracted strong spectral components to efficiently separate the PRFI from the useful signals. By strategically integrating these two approaches, the proposed framework simultaneously exploits the high-intensity and low-rank properties of PRFI for more precise separation, significantly reduces the sensitivity of threshold selection in the WNF process, and facilitates the low-rank conditions required by the FGSVT-LRSD method. Finally, the separated PRFI spectrum is subtracted from the original 2D range spectrum, resulting in the RFI-suppressed SAR image. Experimental results based on both simulated and measured spaceborne SAR data will be presented to demonstrate that, compared to existing methods, the proposed approach exhibits superior PRFI suppression capabilities and effectively preserves useful signals.
- 8 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 32 (10709908) , 2199-2203
Improved SOCP Relaxation for SRI-Unknown Emitter Localization Using a Moving Receiver
Employing a single moving receiver for emitter localization offers numerous advantages, including low cost, ease of implementation, and elimination of clock synchronization. A novel localization method is proposed for a stationary emitter with an unknown Signal Repetition Interval (SRI). We first construct a localization model based on time-of-arrival, taking into account potential missed detections, and then derive the Cramér-Rao Lower Bound. Given that the established model is non-convex, we reformulate the problem into a convex form and convert it to Second-order Cone Programming (SOCP) format. However, the SOCP formulation encounters convex hull issues, which may hinder far-field localization. To address this challenge, we introduce a penalty term to alleviate the convex hull problem for the first time. Simulation results demonstrate the effectiveness of the proposed approach in comparison to existing methods, and the validity of the penalty term is also confirmed.
- 8 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 32 (10709908) , 2199-2203
Improved SOCP Relaxation for SRI-Unknown Emitter Localization Using a Moving Receiver
Employing a single moving receiver for emitter localization offers numerous advantages, including low cost, ease of implementation, and elimination of clock synchronization. A novel localization method is proposed for a stationary emitter with an unknown Signal Repetition Interval (SRI). We first construct a localization model based on time-of-arrival, taking into account potential missed detections, and then derive the Cramér-Rao Lower Bound. Given that the established model is non-convex, we reformulate the problem into a convex form and convert it to Second-order Cone Programming (SOCP) format. However, the SOCP formulation encounters convex hull issues, which may hinder far-field localization. To address this challenge, we introduce a penalty term to alleviate the convex hull problem for the first time. Simulation results demonstrate the effectiveness of the proposed approach in comparison to existing methods, and the validity of the penalty term is also confirmed.
- Top Cited PapersOpen Access
- 11 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 17 (19391404) , 19230-19242
Detection of Pine Wilt Disease Using AAV Remote Sensing With an Improved YOLO Model
Pine Wilt Disease (PWD) is a severe and highly contagious forest disease that poses a significant challenge to sustainable development. This study utilizes Unmanned Aerial Vehicle (UAV) remote sensing and RGB images, combining deep learning techniques to address the challenge of detecting infected trees. The proposed model, YOLO-PWD, integrates the Squeeze-and-Excitation network (SE) and Convolutional Block Attention Module (CBAM) to improve the feature extraction capabilities. Additionally, the integration of a Bidirectional Feature Pyramid Network (BiFPN) enhances the capture of global scene information and adapts to complex environmental variations. We further enhance detection accuracy by incorporating a dynamic convolution kernel, ensuring the system's suitability for deployment on edge devices such as UAVs. Compared to the original YOLOv5s model, YOLO-PWD demonstrates a significant improvement in recall rate, with an increase of 14.2%, and achieves an impressive Average Precision (AP) of 95.2% for detecting discolored pine trees. The precision, recall, and AP for dead pine trees are also enhanced, with precision increasing by 9.7% and AP by 6.7%. Despite a mere 0.9MB increase in model size, the F1-score for discolored pine trees improves by 4.1%, and the F1-score for dead pine trees increases by 3.3%. The experimental results suggest that the YOLO-PWD model can better meet the requirements of PWD detection by UAV remote sensing in large epidemic areas. This advancement carries significant implications for the protection of pine forest resources and contributes to environmentally sustainable development.
- 11 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 17 (19391404) , 19230-19242
Top Cited PapersOpen AccessDetection of Pine Wilt Disease Using AAV Remote Sensing With an Improved YOLO Model
Pine Wilt Disease (PWD) is a severe and highly contagious forest disease that poses a significant challenge to sustainable development. This study utilizes Unmanned Aerial Vehicle (UAV) remote sensing and RGB images, combining deep learning techniques to address the challenge of detecting infected trees. The proposed model, YOLO-PWD, integrates the Squeeze-and-Excitation network (SE) and Convolutional Block Attention Module (CBAM) to improve the feature extraction capabilities. Additionally, the integration of a Bidirectional Feature Pyramid Network (BiFPN) enhances the capture of global scene information and adapts to complex environmental variations. We further enhance detection accuracy by incorporating a dynamic convolution kernel, ensuring the system's suitability for deployment on edge devices such as UAVs. Compared to the original YOLOv5s model, YOLO-PWD demonstrates a significant improvement in recall rate, with an increase of 14.2%, and achieves an impressive Average Precision (AP) of 95.2% for detecting discolored pine trees. The precision, recall, and AP for dead pine trees are also enhanced, with precision increasing by 9.7% and AP by 6.7%. Despite a mere 0.9MB increase in model size, the F1-score for discolored pine trees improves by 4.1%, and the F1-score for dead pine trees increases by 3.3%. The experimental results suggest that the YOLO-PWD model can better meet the requirements of PWD detection by UAV remote sensing in large epidemic areas. This advancement carries significant implications for the protection of pine forest resources and contributes to environmentally sustainable development.
- 7 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 21 (1545598X) , 1-5
Range-Frequency Domain Iterative Notch Filtering Method for RFI Mitigation in SAR
As the electromagnetic environment increases complexity, electromagnetic signals in the same frequency band will interfere with a synthetic aperture radar (SAR), causing distortion in SAR images. Most notch filters set a fixed threshold and zero notch to achieve radio frequency interference (RFI) suppression, which is not adaptive when the energy level of the RFI varies. In this letter, an iterative notch filtering method based on range-frequency domain outlier detection is proposed. It uses hybrid indicators to distinguish the differences of the range spectrum of the Z-normalized SAR single-look complex (SLC) image and the zero-mean complex Gaussian distribution and obtains an adaptive detection threshold based on spectrum data to detect anomalies. Then, 2-D notch filters are constructed to achieve RFI mitigation. In addition, the proposed method performs the above operations with adaptive thresholds in an iterative manner by setting termination conditions to suppress remaining RFI. Compared with other methods, the proposed method can offer more precise RFI detection and effective RFI mitigation. Experiments based on simulated and real SAR data demonstrate the efficacy of the proposed method.
- 7 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 21 (1545598X) , 1-5
Range-Frequency Domain Iterative Notch Filtering Method for RFI Mitigation in SAR
As the electromagnetic environment increases complexity, electromagnetic signals in the same frequency band will interfere with a synthetic aperture radar (SAR), causing distortion in SAR images. Most notch filters set a fixed threshold and zero notch to achieve radio frequency interference (RFI) suppression, which is not adaptive when the energy level of the RFI varies. In this letter, an iterative notch filtering method based on range-frequency domain outlier detection is proposed. It uses hybrid indicators to distinguish the differences of the range spectrum of the Z-normalized SAR single-look complex (SLC) image and the zero-mean complex Gaussian distribution and obtains an adaptive detection threshold based on spectrum data to detect anomalies. Then, 2-D notch filters are constructed to achieve RFI mitigation. In addition, the proposed method performs the above operations with adaptive thresholds in an iterative manner by setting termination conditions to suppress remaining RFI. Compared with other methods, the proposed method can offer more precise RFI detection and effective RFI mitigation. Experiments based on simulated and real SAR data demonstrate the efficacy of the proposed method.
- Open Access
- 22 November 2023
- Published by MDPI AG in Remote Sensing
- Vol. 15 (23) , 5462
Intelligent Detection and Segmentation of Space-Borne SAR Radio Frequency Interference
Space-borne synthetic aperture radar (SAR), as an all-weather observation sensor, is an important means in modern information electronic warfare. Since SAR is a broadband active radar system, radio frequency interference (RFI) in the same frequency band will affect the normal observation of the SAR system. Untangling the above problem, this research explores a quick and accurate method for detecting and segmenting RFI-contaminated images. The purpose of the current method is to quickly detect the existence of RFI and to locate it in massive SAR data. Based on deep learning, the method shown in this paper realizes the existence of RFI by determining the presence or absence of interference in the image domain and then performs pixel-level image segmentation on Sentinel-1 RFI-affected quick-look images to locate RFI. Considering the need to quickly detect RFI in massive SAR data, an improved network based on MobileNet is proposed, which replaces some inverted residual blocks in the network with ghost blocks, reducing the number of network parameters and the inference time to 6.1 ms per image. Further, this paper also proposes an improved network called the Smart Interference Segmentation Network (SISNet), which is based on U2Net and replaces the convolution of the VGG blocks in U2Net with a residual convolution and introduces attention mechanisms and a modified RFB module to improve the segmentation mIoU to 87.46% on average. Experiment results and statistical analysis based on the MID dataset and PAIS dataset show that the proposed methods can achieve quicker detection than other CNNs while ensuring a certain accuracy and can significantly improve segmentation performance under the same conditions compared to the original U2Net and other semantic segmentation networks.
- 22 November 2023
- Published by MDPI AG in Remote Sensing
- Vol. 15 (23) , 5462
Open AccessIntelligent Detection and Segmentation of Space-Borne SAR Radio Frequency Interference
Space-borne synthetic aperture radar (SAR), as an all-weather observation sensor, is an important means in modern information electronic warfare. Since SAR is a broadband active radar system, radio frequency interference (RFI) in the same frequency band will affect the normal observation of the SAR system. Untangling the above problem, this research explores a quick and accurate method for detecting and segmenting RFI-contaminated images. The purpose of the current method is to quickly detect the existence of RFI and to locate it in massive SAR data. Based on deep learning, the method shown in this paper realizes the existence of RFI by determining the presence or absence of interference in the image domain and then performs pixel-level image segmentation on Sentinel-1 RFI-affected quick-look images to locate RFI. Considering the need to quickly detect RFI in massive SAR data, an improved network based on MobileNet is proposed, which replaces some inverted residual blocks in the network with ghost blocks, reducing the number of network parameters and the inference time to 6.1 ms per image. Further, this paper also proposes an improved network called the Smart Interference Segmentation Network (SISNet), which is based on U2Net and replaces the convolution of the VGG blocks in U2Net with a residual convolution and introduces attention mechanisms and a modified RFB module to improve the segmentation mIoU to 87.46% on average. Experiment results and statistical analysis based on the MID dataset and PAIS dataset show that the proposed methods can achieve quicker detection than other CNNs while ensuring a certain accuracy and can significantly improve segmentation performance under the same conditions compared to the original U2Net and other semantic segmentation networks.