Ship detection in SAR images using derivative based methods

Synthetic Aperture Radar (SAR) imagery is widely utilized in maritime applications due to its capability to provide high-quality images regardless of weather and illumination conditions. However, ship detection in SAR images remains a challenging task because of coherent speckle noise, complex coastal backgrounds, and densely distributed targets. In this thesis, both mathematical model-based and deep learning-based approaches are developed for ship detection in low-contrast, inshore, and densely populated SAR scenes. In the proposed mathematical approach, the Hessian matrix and its eigenvalues are first computed to suppress land regions and reduce the effects of speckle noise. The largest eigenvalue is then used as an input to a Gaussian function, and standard deviation and mean maps are generated from the SAR image. These maps are combined using a probabilistic framework to produce an enhanced image that highlights potential ship regions. Finally, morphological operations and connected component analysis are applied to identify ship candidates. Experimental results demonstrate that the proposed method achieves both high detection accuracy and computational efficiency. In addition, YOLO-based deep learning models are employed for SAR ship detection, and extensive experiments are conducted on different datasets. The obtained results indicate that the deep learning approach provides high detection performance and robust ship localization capabilities. Overall, the findings of this thesis demonstrate that both mathematical image processing techniques and deep learning models can effectively address the SAR ship detection problem and provide accurate and reliable results under challenging imaging conditions.