Abstract
Independent of daylight and weather conditions, synthetic aperture radar (SAR) images have been widely used for ship monitoring. The traditional methods for SAR ship detection are highly dependent on the statistical models of sea clutter or some predefined thresholds, and generally require a multi-step operation, which results in time-consuming and less robust ship detection. Recently, deep learning algorithms have found wide applications in ship detection from SAR images. However, due to the multi-resolution imaging mode and complex background, it is hard for the network to extract representative SAR target features, which limits the ship detection performance. In order to enhance the feature extraction ability of the network, three improvement techniques have been developed. Firstly, multi-level sparse optimization of SAR image is carried out to handle clutters and sidelobes so as to enhance the discrimination of the features of SAR images. Secondly, we hereby propose a novel split convolution block (SCB) to enhance the feature representation of small targets, which divides the SAR images into smaller sub-images as the input of the network. Finally, a spatial attention block (SAB) is embedded in the feature pyramid network (FPN) to reduce the loss of spatial information, during the dimensionality reduction process. In this paper, experiments on the multi-resolution SAR images of GaoFen-3 and Sentinel-1 under complex backgrounds are carried out and the results verify the effectiveness of SCB and SAB. The comparison results also show that the proposed method is superior to several state-of-the-art object detection algorithms.
Highlights
Due to the all-weather, all-day characteristics, synthetic aperture radar (SAR) has become one of the important means of earth observation [1], such as vehicle detection [2], river detection [3] and image recognition [4,5]
In order to verify the effectiveness of the proposed Split Convolution Block (SCB) and Spatial Attention Block (SAB), we conduct the following comparison experiments under four conditions: RetinaNet, SAB-RetinaNet (RetinaNet embedded with SAB), SCB-RetinaNet (RetinaNet embedded with SCB), 2S-RetinaNet (RetinaNet embedded with SAB and SCB)
In order to verify the superior performance of the method we proposed, we compare the performance of our method with three other state-of-the-art object detection methods: Faster R-convolution neural network (CNN) [45], SSD [55], YOLOv3 [56]
Summary
Due to the all-weather, all-day characteristics, SAR has become one of the important means of earth observation [1], such as vehicle detection [2], river detection [3] and image recognition [4,5]. The traditional method for ship detection from SAR images includes statistical distribution-based methods [6,7,8,9], multi-scale-based methods [10], template matching [11] and multiple/full polarization-based methods [12,13]. These methods highly rely on the distributions of sea clutters and the predefined thresholds [14,15,16]. Deep learning has been applied to SAR ship detection [19,20,21,22,23,24,25]
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
