Ship detection in Synthetic Aperture Radar (SAR) imagery plays a pivotal role in maritime surveillance, national security, and navigation safety. Conventional model such as Constant False Alarm Rate (CFAR) algorithms are constrained by sea clutter modeling and parameter estimation, often resulting in reduced detection accuracy under complex oceanic conditions. This paper presents an inclusive survey of Neural Network methods for SAR ship detection, reviewing available datasets, detection architectures, and frameworks. Furthermore, we experimentally evaluate state-of-the-art models, including ResNet-18, VGG16 and YOLOv8 on widely used datasets such as SSDD and SAR-Ship-Dataset. The results demonstrate the superior accuracy and efficiency of YOLOv8 ResNet-18 compared to VGG16, highlighting the advantages of residual learning in high-precision maritime applications.