Benefiting from the rapid development of deep learning, the field of synthetic aperture radar (SAR) ship detection has been promoted with renewed vigor. However, due to the unique characteristics of SAR ship detection, several challenges have been encountered in the process of fusing SAR ship detection. On the one hand, some special properties of SAR images, including low resolution, small targets, dense ship inshore arrangements, and background clutter noise, increase the difficulty of detection; on the other hand, satellite detection requires a high degree of real-time and lightweight, and most traditional models fail to satisfy the requirements in two areas. After considering the above issues, we combine the advantages of space-to-depth convolution (SPDConv) and inception depthwise convolution (InceptionDWConv) to design SPD-InceptionDWConv (SIDConv), which resolves the above problems with fewer parameters and operations. Moreover, based on SIDConv, we also design light SIDConv (LSIDConv) and efficient SIDConv (ESIDConv) to further reduce the model complexity and to accelerate the detection process. With the above module as the core and YOLOv5 as the base, we propose a lighter and faster detector named LFer-Net, which has only 0.8M parameters and 1.9G FLOPs. Extensive experimental results on the SSDD and HRSID datasets confirm that our model is not only superior to other models with 98.2 and 90.6 accuracy but also achieves detection speeds of 144 and 96 frames per second (FPS), respectively.
To achieve high-resolution and wide-swath (HRWS) imaging, a space-time waveform-encoding (STWE) spaceborne synthetic aperture radar (SAR) system is adopted. In rugged terrain, the beam-pointing mismatch problem will appear when the traditional digital beamforming (DBF) technique is used to separate the received echoes. This problem leads to decreasing the received echo’s gain, deteriorating the quality of the image product and making the interpretation of SAR image difficult. To address this problem, an advanced echo separation scheme for STWE spaceborne SAR based on the DBF and blind source separation (BSS) is put forward in this paper. In the scheme, the echoes are transmitted within the adjacent pulse repetition intervals to simulate multiple beams, and the scattered echoes are received by the sixteen-channel antennas in elevation simultaneously. In post-processing, a detailed flow is adopted. In the method, the DBF is firstly performed on received echoes. Due to the error caused by terrain undulation, the degree of echo separation is not enough. Then, the BSS is performed on the multiple echoes obtained after the DBF processing. Finally, the highly separated echo signal can be obtained. In this scheme, there is no need to perform the direction of arrival (DOA) estimation before the DBF processing, which saves valuable computing resources. In addition, to verify the proposed scheme, point target and distributed target simulations based on the 16-channel data of an elevation X-band DBF-SAR system are carried out. The results of point targets indicate that the residual echo caused by rough terrain can be reduced by more than 14 dB using the proposed scheme. The proposed scheme can be directly implemented into existing SAR systems; thus, it does not increase the complexity of the system design. The scheme has the potential to be applied to future spaceborne SAR missions.
12 January 2021 in IEEE Geoscience and Remote Sensing Letters
The 3-D structure of forests is an important indicator for evaluating forest health and can provide data support for ecological monitoring and protection. Synthetic aperture radar (SAR) tomography (TomoSAR) is an important technology for forest structure estimation using the multibaseline (MB) SAR data stacks. The spectral estimators, such as the Capon method, are usually used to estimate the 3-D reflectivity along elevation direction. In this letter, a modified Capon (M-Capon) method is proposed to improve the estimated profiles of ground and canopy scatterers in the elevation direction, including the estimation accuracy and resolution. This method combines the ideas of the CLEAN algorithm with the Capon method and uses iteration to improve the resolution and accuracy of the estimation. The effectiveness of the M-Capon method is demonstrated using the simulated data and the MB SAR data acquired by the P-band airborne SAR system over the Saihanba Forest Farm in Hebei, China.
19 January 2024 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
With the expanding scope of human activities in marine environments, the efficient detection and tracking of mobile targets on the ocean's surface have become increasingly crucial. SAR constellation can obtain ground observation data based on user requests and subject to visibility conditions. Now it is an indispensable tool in sea surface moving target search Tasks. Satellite constellation resources are scarce and limited, and user demands are diverse. How to rationally dispatch satellite constellation resources to meet user needs to the maximum extent and improve the application efficiency of satellite resources is an urgent scientific problem that needs to be solved. This paper mainly expounds two respects of work: firstly, modeling SAR constellation scheduling problem for sea surface moving target search tasks to establish the objective function; secondly, a novel multi-strategy discrete constrained differential evolution algorithm denoted as MSDCDE is proposed in the paper. The proposed MSDCDE algorithm integrates cross strategy based on discrete variables, constraint handling techniques, population restart strategy, and left-shift local strategy, which can effectively avoid falling into local optimality, thereby achieving global optimality and improving search and rescue performance. Six sets of experiments, totaling 215 runs, have been conducted to validate the effectiveness of the proposed resolution process framework and the MSDCDE algorithm. The proposed method demonstrated an over 48.98% performance improvement compared to some state-of-the-art algorithms and significantly reduced task completion time.
4 August 2021 in IEEE Transactions on Geoscience and Remote Sensing
The innovative spaceborne Earth observation mission LuTan-1 (LT-1) deploys advanced full polarimetric L-band synthetic aperture radar (SAR) to obtain high-precision, multidimensional ground feature information. As the quad-pol SAR system, LT-1 suffers from strong ambiguities that degrade the quality of the observation products. Antenna pattern synthesis algorithm can suppress the range ambiguities without the increase of the azimuth ambiguities and the system complexity and therefore has great potential to improve the overall ambiguity performance of the quad-pol SAR system. However, the previous research on this kind of method is generally limited to specific situations and lacks of the analysis of actual measurement results. Based on the application requirements of LT-1, this article proposes a novel pattern synthesis algorithm that suppresses the range ambiguities via sequential convex optimizations. In simulation and comparison, the proposed algorithm effectively suppresses the strong co-polarized range ambiguities and shows the flexibility, efficiency, and stability that are significantly better than the previous algorithms. What is more, the analysis of practical performance loss is performed based on the measurement result of the LT-1 antenna, and the practicality and validity of the algorithm are strongly verified.