基于深度卷积神经网络的SAR舰船目标检测

针对传统合成孔径雷达(synthetic aperture radar,SAR)图像舰船目标检测算法检测精度易受斑点噪声影响,且只能提取底层特征及其泛化性较差的问题,提出了一种基于深度卷积神经网络的SAR图像舰船目标检测算法。首先将目前先进的单次多盒检测器(single shot multibox detector,SSD)检测算法应用到SAR图像舰船目标检测领域,指出了其在该领域存在的局限性,在此基础上提出了基于SSD的新的检测方法,包括融合上下文信息,迁移模型学习,在公开的SSDD数据集上进行了训练和测试,对实验结果进行了对比分析,实验结果表明,相比于原始的SSD检测算法,所提出的方法不仅提高了目标检测精度,同时也保证了算法的检测效率。

Ship detection in SAR images based on deep convolutional neural network

Aiming at the problems that the detection accuracy of ship detection in synthetic aperture radar (SAR) images is susceptible to speckle noise, and traditional algorithms only extract the underlying features and the generalization is poor, this paper proposes a ship detection algorithm based on deep convolutional neural network. First, the current single shot multibox detector (SSD) detection algorithm is applied to the field of SAR image ship detection, and its limitations are pointed out. Next, improved detection methods based on SSD are proposed, including context information fusion and transfer model learning. Finally, the experimental results on the open ship dataset show that, compared with the original SSD detection algorithm, the proposed method not only improves the target detection accuracy, but also ensures the efficiency of the algorithm.