A Sonar Image Recognition Method Based on Convolutional Neural Network Trained through Transfer Learning
-
摘要: 针对利用传统训练方法进行声呐图像识别时缺乏数据的问题, 文中提出一种利用迁移学习训练卷积神经网络(CNN)实现声呐图像识别的方法。基于迁移学习的原理, 通过对已有的预训练网络进行微调与重新训练, 以减小对训练数据量的需求。随后利用缩比模型水池试验验证了该方法的有效性。试验结果表明, 基于AlexNet预训练网络, 相比传统的学习方法, 迁移学习方法可以利用较少的训练数据, 在较短的时间内通过训练达到95.81%的识别率。试验还对比了基于6种预训练网络进行迁移学习后的网络性能, 结果表明基于VGG16的迁移网络识别率最高, 可达到99.48%。最后, 试验结果验证了CNN具有一定的噪声容忍能力, 在较强噪声背景下, 能保证较高的识别率。Abstract: A sonar image recognition method using convolutional neural network trained by transfer learning is proposed aiming at the data shortage problem in making use of conventional training method for sonar image recognition. Based on the principle of transfer learning, the existing pre-trained network is fine-tuned and retrained to reduce the demand for training data. Pool experiment of the scale model verifies the effectiveness of this method. Experimental results show that, on the basis of AlexNet pre-trained network, the transfer learning method uses less training data compared with the conventional learning method, and achieves a recognition rate of 95.81% in a short time. The experiment also compares the performance of the network after the transfer learning based on six kinds of pre-trained networks, and the result shows that the recognition rate of the transfer network based on VGG16 is the highest, which can reach 99.48%. It is verified that the deeply convolutional neural network has certain noise tolerance ability, and can ensure a high recognition rate in the background of strong noise.
-
[1] 朱兆彤, 彭石宝, 许稼, 等. 水下目标亮点拓扑特征提取及自动识别方法[J]. 声学学报, 2018, 43(2): 154-162.Zhu Zhao-tong, Peng Shi-bao, Xu Jia, et al. Multiple Highlights Topology Vector Feature Extraction and Automatic Recognition Method for Underwater Target[J]. Acta Acustica, 2018, 43(2): 154-162. [2] Krizhevsky A, Sutskever I, Hinton G E. ImageNet Classification with Deep Convolutional Neural Networks[J]. Communications of the ACM, 2017, 60(6): 84-90. [3] He K, Zhang X, Ren S, et al. Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Clas-sification[C]//2015 IEEE International Conference on Computer Vision(ICCV). Santiago, Chile: IEEE, 2015: 1026-1034. [4] Szegedy C, Liu W, Jia Y, et al. Going Deeper with Convolutions[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Boston, Massachusetts, USA: IEEE, 2015: 1-9. [5] Hu J, Shen L, Sun G. Squeeze-and-Excitation Net-works[EB/OL]. (2017-09-05)[2018-09-18]. http://adsabs. harvard.edu/abs/2017arXiv170901507H. [6] McKay J, Gerg I, Monga V, et al. What’s Mine is Yours: Pretrained CNNs for Limited Training Sonar ATR[C]//Oceans 2017-Anchorage. Alaska, USA: Oceans, 2017: 1-7. [7] Gao F, Huang T, Wang J, et al. Dual-Branch Deep Convolution Neural Network for Polarimetric SAR Image Classification[J]. Applied Sciences, 2017, 7(5): 447-465. [8] Huang Z, Pan Z, Lei B. Transfer Learning with Deep Convolutional Neural Network for SAR Target Classification with Limited Labeled Data[J]. Remote Sensing, 2017(9): 907-927. [9] Glorot X, Bordes A, Bengio Y. Deep Sparse Rectifier Neural Networks[C]//International Conference on Artificial Intelligence and Statistics. Ft. Lauderdale, UK, 2011: 315-323. [10] 卢建斌, 张云雷, 席泽敏, 等. 基于统计特征的水下目标一维距离像识别方法研究[J]. 声学技术, 2015, 34(2): 121-126.Jianbin Lu, Yunlei Zhang, Zemin Xi et al., Research on underwater target recognition with statistical features of high resolution range profiles [J]. Technical Acoustics, 2015, 34(2): 121-126. [11] 闫慧辰, 彭石宝, 许稼, 等. 水下旋转目标缩比声纳成像方法及系统: CN104297756A[P]. 2014-10-09.
点击查看大图
计量
- 文章访问数: 711
- HTML全文浏览量: 2
- PDF下载量: 332
- 被引次数: 0