Abstract: With the development and application of the high resolution synthetic aperture radar (SAR) system, the rapid and accurate recognition and classification of ship targets based on SAR images has become an important means of maritime target reconnaissance and surveillance. A new method for geometric structure feature of ship targets in SAR images based on fine segmentation is proposed. Firstly, the Radon transform is adopted to separate ship targets with the "ghosting" or the "spider". The primary area of ship targets is obtained by the image threshold segmentation method, image processing based on the morphological method is adopted to inhibit the side lobe. Secondly, the target zone is refined based on the method of elliptical shape constraint, the "burr" phenomenon and the regional fracture phenomenon are well solved, and then the best image segmentation area for ship targets is acquired. Finally, the minimum circumscribed rectangle of target zone is obtained based on the successive approximation method to extract the geometric structure feature accurately. The experiments on the obtained GF-3 satellite SAR images demonstrate that our method proposed in this paper to extract the geometric structure feature of ship targets with high accuracy and strong stability, and it is significant for the identification and classification of ship targets at sea.