样本仿真结合迁移学习的声呐图像水雷检测

水雷检测对于国防安全具有重要意义，然而，由于水下目标声呐成像实验代价较大，通常难以获得足够的水雷声呐图像样本，因此导致难以应用深度神经网络提高水雷等目标的检测精度。针对这一问题，提出样本仿真结合迁移学习的侧扫声呐图像水雷目标检测与识别方法。首先，根据侧扫声呐成像机理，建立水雷目标的仿真模型，进而仿真得到大量水雷目标样本；然后，采用大型广源域数据集ImageNet对深度卷积神经网络进行预训练，再用真实水雷样本和仿真水雷样本对深度卷积神经网络进行微调以适应水雷目标；最后，将微调后的深度卷积神经网络作为目标检测的基准网络，并进行目标检测训练；采用真实的水下水雷声呐图像数据对训练完成的网络进行验证和比较。实验结果表明，提出的基于样本仿真和迁移学习的侧扫声呐图像水雷目标检测方法能够更好地检测水雷目标，优于传统的特征提取及检测方法及只采用真实样本进行训练的检测方法，对于水下目标检测具有借鉴意义。

Detection of underwater mine target in sidescan sonar image based on sample simulation and transfer learning

Mine detection is of great significance to national defense and security. However, because of the high cost of underwater target sonar imaging experiments, usually sufficient mine sonar image samples cannot be obtained. Thus, the detection accuracy of mines and other targets is difficult to improve using deep neural networks. To solve this problem, a mine target detection and recognition method based on image simulation and transfer learning is proposed in this paper. The simulation model of a mine target is established in this proposed method according to the imaging mechanism of sidescan sonar to obtain numerous mine target images. Further, the deep convolutional neural network (CNN) is pretrained by the ImagenetNet dataset and then the deep CNN is fine-tuned by real and simulated mine images to adapt to mine targets detection. Finally, the fine-tuned CNN is used as the benchmark of the target detection network, and the target detection training is performed; real mine sonar images are used to verify the effectiveness of the proposed method. Experimental results show that the proposed mine detection method based on sample simulation and transfer learning can detect mine targets more accurately than some traditional feature extraction and detection, and deep learning-based detection methods using only real samples for training, which is helpful for underwater target detection.