SAR旋转船舶检测识别数据集

深度学习在SAR舰船探测领域得到了广泛的应用。然而，当前SAR舰船探测仍面临场景复杂、尺度多、目标小等诸多挑战。为了促进解决上述问题，本文发布了高分辨率SAR舰船探测数据集，可用于旋转帧目标检测。该数据集包含六类船舶。总共将30个分辨率为1 m的中国高分三号全景SAR图块裁剪成切片，每片1024×1024像素。此外，数据集中的大多数图像都包含具有复杂背景干扰的近岸区域。使用八个最先进的旋转探测器和基于CFAR的方法对数据集进行评估。实验结果表明，复杂的背景将对探测器的性能产生很大影响。

Deep learning has been widely applied in the field of SAR ship detection. However, current SAR ship detection still faces numerous challenges including complex scenes, multi-scale variations, and small-sized targets. To address the aforementioned issues, this paper releases a high-resolution SAR ship detection dataset for rotating frame object detection. This dataset covers six ship categories. A total of 30 full-scene SAR tiles from China's Gaofen-3 (GF-3) with 1 m resolution are cropped into 1024×1024 pixel slices. Moreover, most images in the dataset contain nearshore regions with complex background interference. The dataset is evaluated using eight state-of-the-art rotating detectors and CFAR-based methods. Experimental results demonstrate that complex backgrounds can significantly impact the performance of detectors.

搜集汇总

背景与挑战

背景概述

该数据集是一个高分辨率SAR旋转船舶检测数据集，专门用于旋转框目标检测任务，包含六类船舶，数据来源于高分三号卫星聚束模式，分辨率为1米，图像切片尺寸为1024×1024像素。数据集特点在于近岸场景占比高（63.1%），背景复杂干扰显著，旨在评估和提升SAR船舶检测算法在复杂环境下的性能。