融合注意力机制与改进SSD算法的SAR舰船目标检测方法

针对单次多盒检测算法(SSD)对复杂背景下合成孔径雷达(SAR)图像舰船目标的检测容易出现误检或漏检情况,提出一种基于融合注意力机制与改进的SSD算法的目标检测方法。首先在SSD算法上引入ResNet网络并进行改进,以提供丰富的语义信息和细节信息,提高算法的鲁棒性;其次融合通道和空间注意力增强对舰船目标的辨认能力,抑制海杂波等干扰信息;同时改进损失函数来解决舰船密集分布时的漏检问题,提高网络训练效果。数据集上的实验表明,该方法平均准确率(mAP)为87.6%,比SSD算法提高了4.2个百分点,目标的漏检和误检明显减少。相比SSD算法,该算法对复杂背景下的舰船目标有较好的辨别能力和鲁棒性,抗干扰能力有所提升。

SAR ship object detection method incorporating attention mechanism and improved SSD algorithm

This paper proposed a object detection method based on a fused attention mechanism with an improved single shot multibox detector(SSD) algorithm for the detection of synthetic aperture radar(SAR) image ship targets in complex backgrounds by SSD original algorithm that was prone to false detection or missed detection. Firstly, this method introduced the ResNet and improved SSD algorithm to provide rich semantic and detailed information to improve the robustness. Secondly, the fusion of channel and spatial attention enhanced the recognition ability of ship targets and suppressed the interference information such as sea clutter. Meanwhile, it improved the loss function to solve the problem of missed detection when the ships were densely distributed and to improve the network training effect. Experiments on the dataset show that the mean average precision(mAP) of the method is 87.6%, which is 4.2 percentage points higher than that of the SSD algorithm, and the missed and false detections of targets are significantly reduced. Compared with the SSD algorithm, this method has better discrimination ability and robustness for ship targets in complex backgrounds, and improves the anti-interference ability.