H2Det: A High-Speed and High-Accurate Ship Detector in SAR Images

Abstract

Synthetic aperture radar (SAR) sensor is a vital platform for ship detection whose accuracy and speed are usually difficult to balance. An urgent problem to be solved is how to achieve high-speed detection while maintaining high-accurate. To address this problem, we propose a high-speed and high-accurate detector (H2Det) in SAR images. For one thing, we adopt fewer convolutional layers, CSP module and rectangle filling to ensure model high-speed. For another, we propose spatial pyramid pooling (SPP), bottom-up path augmentation (PA), and mosaic data augmentation to ensure model high-accurate. To establish an optimal H2Det, we conduct comparative studies on SSDD dataset. Moreover, we verify the effectiveness of these modules mentioned above through ablation studies. The experimental results on SAR ship detection dataset (SSDD) demonstrate that both accuracy and speed of the proposed method outperform other state-of-the-art methods and references. In addition, the strong migration ability of the proposed H2Det is shown on high-resolution SAR images dataset (HRSID).

A Precise Oriented Ship Detector in SAR Images Based on Dynamic Rotated Positive Sample Mining

“…The convolution neural networks are adapted from common objects detection to SAR ship detection. Zhu et al [6] inserted spatial pyramid pooling into the architecture of Tingxuan Yue, Yanmei Zhang, Jin Wang, Yanbing Xu, Pengyun Liu and Chengcheng Yu are with Beijing Institute of Technology, Beijing 100081, China.…”

“…The adverse effects of fuzzy areas caused by SAR imaging mechanism have been reduced by guide position regression branch and redesigned regression method. The mosaic data augmentation was introduced to solve the limit of insufficient data in the stage of pre-process ship samples [6]. These structures and tactics designed for detecting ships can locate ships more accurately.…”

SPANet: A Self-Balancing Position Attention Network for Anchor-Free SAR Ship Detection

“…One is the detection algorithm that introduces the denoising algorithm as a preprocessing module [20], [21], [22]. The other is the optimization algorithms for SAR target characteristics, directly design dedicated processing module, such as attention reception pyramid [23], cascaded multidomain attention [24], continuous attention module [25], AFSar [26], BANet [27], and H2Det [28].…”

“…Therefore, the problem of mining image semantic features in complex scenes urgently needs to be solved. 3) To break through the bottleneck of SAR image detection, researchers have designed specialized processing modules for various characteristics of SAR images with complex background, multi-scale targets, irregular distribution, respectively [23]- [28]. However, such distributed networks lack integrated design and tend to have more repetitive operations, resulting longer processing times for reasoning and training.…”

A Ship Detection and Imagery Scheme for Airborne Single-Channel SAR in Coastal Regions

“…In recent years, some ship detection algorithms from SAR imagery based on deep learning have been widely applied [15][16][17][18][19][20][21][22]. Deep learning is a trend of ship detection from SAR images in the future, and has made significant progress, but it usually conforms to the optical ship detector (OSD) and the postprocessing of spaceborne SAR images, and depends on a high resolution or on complex computation, which makes it not suitable for a real-time SAR system such as airborne SAR [23][24][25]. For a single-channel airborne SAR, a ship detection system requires realtime processing; constant false alarm rate (CFAR) detection is a good choice, but a single CFAR cannot be directly performed for SAR images [26].…”