11 February 2025 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Maritime ship detection is an essential prerequisite for targeting key objectives and safeguarding the security of territorial waters. The all-day and all-weather operational capability of Synthetic Aperture Radar(SAR) makes SAR-based ship detection indispensable for maritime surveillance. Marine surveillance requires extensive coverage for large-scale searches. But there is a contradiction between high resolution and wide swath, which cannot be taken into account. Most existing marine surveillance modes generally exchange extensive coverage at the expense of spatial resolution. The two-dimensional(2D) detailed information is lost in this way, which is not conducive to target classification. Furthermore, real-time detection becomes a challenge under the constraints of limited on-board conditions. In this article, a novel marine surveillance mode is proposed. It divides the target search task into two parts: wide-area target detection in 2D asymmetric resolution mode(2D-ARM) and key target focusing in 2D high-resolution mode. In the context of 2D-ARM, the sparsity of the targets and the joint sparse low-rank characteristics of the background are studied, and a low-rank approximation model is developed to accomplish the task of real-time ship detection. The experiments show that the proposed method can realize the sparse target matrix extraction(STME) through matrix decomposition. Using Sentinel1A, ALOS PALSAR and Gaofen3 single-polarization single-look complex(SLC) data, the 2D-ARM imaging is simulated, and a dataset is built to verify the processing performance of the proposed method.
26 February 2024 in IEEE Geoscience and Remote Sensing Letters
This letter proposes a scheme for automatically and accurately extracting the parameters of weak radio frequency interference (RFI) present in the signal within the time-frequency domain. This scheme utilizes Short-Time Fourier Transform (STFT) to transform each frame of radar echo containing interference into the time-frequency domain. In the time-frequency domain, a joint low-rank sparse robust principal component analysis based on truncated nuclear norms (TJLS-RPCA) is employed to divide the time-frequency representation into two joint low-rank sparse matrices, from which the interference parameters are extracted. Subsequently, K-means clustering is applied to determine the number of RFIs, and Gaussian fitting is applied to eliminate high-error items. This scheme can extract parameters such as the central frequency, bandwidth, and duration of the interference. The proposed scheme improves the accuracy of parameter extraction and processing speed, demonstrating notable effectiveness in handling weak interference. The Monte Carlo experimental results demonstrate that the scheme provides highly accurate parameter estimation when the Signal-to-Interference Ratio (SIR) is less than 13dB, with errors below 0.5%. Finally, the feasibility of the proposed scheme is validated using Sentinel-1A data, and the extracted results are consistent with the simulation results.
19 April 2024 in IEEE Geoscience and Remote Sensing Letters
In the above article [1] , there is a correction to the author list. The author list is as follows:
29 September 2025 in IEEE Geoscience and Remote Sensing Letters
Synthetic aperture radar (SAR), capable of ultra-wide swath imaging, serves as a vital tool for ocean surveillance. Ship detection in SAR imagery holds significant importance for maritime security and traffic management. Current ship detection algorithms exhibit excessive computational complexity, and both conventional and deep learning-based approaches struggle to achieve rapid detection in ultra-wide swath SAR imagery acquired by next-generation SAR systems. This paper proposes a rapid adaptive anchor framework for ship detection. First, the detection criterion statistics are decomposed into multiple prefix sum matrices, ensuring minimal computational complexity for anchor boxes of arbitrary sizes. Second, the image is quadrant-divided, with detection criterion statistics computed separately to determine ship presence in each quadrant. Third, quadrants containing ships undergo iterative quadrant division, while ship-free quadrants are annotated using differential matrices. Finally, the initial anchor boxes are systematically shifted to repeat the second and third steps across the entire image, ultimately generating a sea clutter mask. The computational complexity of the proposed ship detection framework remains equivalent to that of loading the full SAR image, introducing negligible additional overhead to existing systems. The framework is validated using empirical SAR datasets, demonstrating enhanced computational efficiency without compromising detection accuracy.
9 October 2025 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
In recent years, the deep learning-based synthetic aperture radar (SAR) ship detection technology has been rapidly developed, and has been widely used in fishery management, maritime surveillance, maritime rescue and other fields, but there are still some challenges. For instance, the feature maps extracted by the backbone network contain substantial redundant features, and repeatedly processing these redundant features introduces significant computational redundancy and inflates the model's parameter count. Furthermore, the blurred target edges caused by SAR imaging mechanisms also degrade detection accuracy. In order to solve these problems, this paper first proposes a backbone network using multi-receptive field partial convolution (MRFPconv) to extract features. While preserving the quality of backbone feature maps, our approach eliminates significant computational redundancy and reduces the parameter count. Subsequently, a Neck network with Bi-directional Concatenation (BiC) modules and reparameterizable modules, Reparameterizable Bi-directional Path Aggregation Network (RepBi-PAN), was adopted. This architecture reduces network parameters while incorporating shallower features, thereby enhancing the network's detection capabilities. Finally, this paper uses expert experience to design a ship contour extraction method based on image morphology. The extracted contours and a specially designed dual-branch network architecture with a channel attention mechanism are used to further enhance the network's detection capability. A lot of experiments are carried out on the SAR ship detection data sets HRSID and SSDD. MRFP-DBDet achieves a mAP value of 69.3% and an ${F_{1}}$ of 0.895 on the HRSID dataset, and a mAP value of 74.6% and an ${F_{1}}$ of 0.975 on the SSDD dataset, with a model parameter size of only 4.135M. The experimental results demonstrate that the algorithm achieves state-of-the-art detection performance among SAR ship detection methods with comparable parameter counts.
19 December 2025 in IEEE Geoscience and Remote Sensing Letters
Deep learning-based methods hold significant potential for synthetic aperture radar (SAR) target detection, but they still face numerous challenges, including difficulty extracting global contextual features for large-scale targets, significant multi-scale issues, and the problem of feature extraction of SAR targets with large aspect ratios, which hinder further performance improvement. To this end, this paper proposes a WaveFormer module, which decomposes the image through wavelet convolution and uses convolution and Transformer to process the frequency domain components they are good at, respectively, to expand the receptive field with low parameter overhead and enhance the target feature extraction ability. To address cross-layer information attenuation during feature fusion, a Gather-and-Distribute(GD) mechanism is introduced to reconstruct the Neck network, enhancing multi-scale feature fusion and detection capabilities. Furthermore, given the large aspect ratio and distinct principal axis orientation of SAR targets, a Weighted Strip-Convolution(WSConv) is proposed to effectively improve detection performance. Experiments on the largest multi-class SAR target detection dataset, SARDet-100K, demonstrate that our method achieves a mean average precision (mAP) of 61.5%, reaching state-of-the-art performance and validating its effectiveness.