基于似圆阴影的光学遥感图像油罐检测

针对光学遥感图像中受阴影干扰的油罐目标识别率低的问题，该文提出一种将改进的视觉显著模型与似圆阴影区域特征检测相结合的由粗到精的油罐目标检测方法。首先建立改进的视觉显著模型，将油罐从复杂背景中粗分离。然后对分离结果中由油罐产生的似圆阴影区域进行精检测，得到疑似油罐目标。再去除阴影，获得油罐目标的初步检测结果。最后基于图搜索策略及先验知识，确定油罐目标并定位油库区域。实验结果表明，该方法对检测光学遥感图像中存在似圆阴影的油罐目标具有较高的鲁棒性和准确率。同时，在不同环境的光学遥感图像中使用该方法可快速准确地定位油库区域。

Oil Tank Detection in Optical Remote Sensing Imagery Based on Quasi-circular Shadow

To deal with the issue of low oil tanks recognition rate in optical remote sensing image, an improved oil tanks detection method is proposed, which is based on the improved visual saliency model and quasi-circular shadow region. Firstly, the oil tanks are separated from the complex background by using the improved visual saliency model. Secondly, the circular shadow regions are finely detected, and the suspected oil tanks are obtained. Then, the shadow region and the preliminary detection result of oil tanks are obtained. Finally, the false oil tank targets are removed and oil depots are determined based on graph search strategy and prior knowledge. The proposed method is robust to the oil tanks in the optical remote sensing images, and can effectively detect the oil tanks in high recognition rate. The experimental results indicate that the proposed algorithm are fast and accurate to detect the oil tanks, which is suitable for optical remote sensing images of different spatial resolutions.