Research On Ship Detection Algorithm In Synthetic Aperture Radar Image Based On Multi-Scale Feature Fusion

Synthetic Aperture Radar(SAR)is a type of Earth observation technology that operates actively and is capable of capturing images of the Earth regardless of the weather conditions.It can be installed on a variety of platforms,including aircraft,satellites,and spacecraft,to enable round-the-clock monitoring of the Earth’s surface.With the successful operation of Terra SARX,RADARSAT-2,Sentinel-1,Gaofen-3,and other orbiting satellites,has resulted in the widespread utilization of SAR for various purposes such as military surveillance,disaster monitoring,environmental assessment,and maritime management.The continuous improvement of radar technology has enabled SAR images to achieve higher and higher resolutions,which makes it possible to use SAR images to detect and identify marine targets with high accuracy.In recent years,artificial intelligence technology represented by deep learning has developed rapidly.Deep learning is an algorithm for constructing deep networks to learn the inherent laws and representation levels of sample data.The vigorous development of deep learning provides new ideas and methods for SAR ship target detection.However,due to the complexity of microwave imaging mechanisms,SAR ship target detection is still a challenging task.In this paper,in-depth research is conducted on SAR ship detection using deep learning techniques.This paper presents an analysis of the challenges and difficulties associated with the detection of ship targets using SAR technology.The limitations of current methods are identified,and proposed enhancements are provided to address the remaining issues.The primary focus of this paper is outlined as follows:(1)In SAR image,most of the regions in the inshore scene include scattered spots and noises,which dramatically interfere with ship detection.Besides,SAR ship images contain ship targets of different sizes,especially small ships with dense distribution.Unfortunately,small ships have fewer distinguishing features making it difficult to be detected.To solve these problems,a SAR ship detection method based on feature enhancement pyramid and shallow feature reconstruction is proposed.This method designs a feature enhancement pyramid,which includes a spatial enhancement module to enhance spatial position information and suppress background noise,and the feature alignment module to solve the problem of feature misalignment during feature fusion.Additionally,to solve the the difficulty of detecting small ship targets in SAR ship images,a shallow feature reconstruction module is designed to obtain the semantic information of small ships.The effectiveness of the proposed approach for detecting ship targets using SAR technology is validated through experiments conducted on two publicly accessible datasets,namely the SAR ship detection Dataset(SSDD)and High-Resolution SAR Images Dataset(HRSID).The results obtained from the experiment demonstrate that the proposed approach is capable of significantly enhancing the accuracy of ship detection in SAR ship detection tasks.(2)Currently,most of SAR ship detection method require pre-setted anchor boxes as prior knowledge.However,the sparsity distribution of ships in SAR images means that most anchor boxes of the detector are redundant,severely reducing the detection efficiency of the detector.And the anchor settings directly affects the performance and generalization ability of the detector.In addition,the difference of ship scales and the strong interference of inshore backgrounds bring significant challenges to the detection performance of SAR ship detectors.To solve the above problems,an anchor-free based SAR ship detection method is proposed.This method adopt an anchor-free based detection strategy to predict bounding boxes avoiding inefficiency of anchor-based detectors.Besides,a global context-guided feature balanced pyramid is designed,which enhances the representation of ship features by balancing multi-level semantic information and learning global context information.Considering the interference of a large amount of scattered noise,a unite attention module was designed to enhance ship features and suppress scattered noise.The experimental results on SSDD and HRSID datasets show that the proposed method achieves better performance compared to other methods.