An Auto-Adapt Multi-Level Threshold Segmentation Method of Ships Detection in Remote Sensing Images with Complex Sea Surface Background

A new multi-level threshold segmentation approach proposed for ship detection. This method is designed to detect targets in remote sensing images, especially for which with complex sea surface background image. An experiment over 1104 ship samples and 11600 no-ship samples, those from Spot, Quickbird, Ikonos, Landsat, shows that the target detection rate of the new method can be as high as 99.5% and the false alarm rate is low. Experiments over images with various content and from different aircraft testify to the new method's robustness.