一种改进的Mask R-CNN卫星影像船舶尾迹检测方法

An improved satellite image ship wake detection method based on Mask R-CNN

为提高运用深度学习算法进行船舶尾迹检测的准确度,本文提出了一种改进的Mask R-CNN网络结构,在传统的Mask R-CNN深度学习算法结构基础上,引入平衡特征金字塔串联结构,增强特征的融合和可辨识性,并引入GCNet提高特征提取能力,改善船舶尾迹的检测效果。以landsat8卫星遥感图像为数据集,通过在不同背景中的船舶航行图像下,比较改进结构与一般Mask R-CNN的检测效果,说明在相同条件下,改进结构较传统的Mask R-CNN算法能够得到更好的检测效果。

In order to improve the accuracy of ship wake detection using deep learning algorithms,this paper proposes an improved Mask R-CNN network structure.Based on the structure of the traditional Mask R-CNN,a series structure of balanced feature pyramids is introduced to enhance the fusion and recognizability of features,and GCNet is incorporated to improve the feature extraction ability.Comparison of detection effects on landset8 dataset show that our proposed improved method achieves better results than the traditional Mask R-CNN.