适用于水下目标识别的快速降维卷积模型
|
摘要
针对传统卷积神经网络在相对较小的数据集上训练容易过拟合的问题,本文提出一个适用于水下目标识别的快速降维卷积网络模型(FRD-CMA)。该模型基于卷积核与特征图对应关系描述模型在数据集上的注意力,并以此进行快速降维,从而降低模型在小数据集上应用时存在的过拟合风险。FRD-CMA模型支持水下目标辐射噪声的端到端处理,通过提取辐射噪声的声音特征并依照水听器的时序关系进行矢量化处理,可以保持模型源输入特征不被破坏。试验结果表明:相较于之前的水下目标识别任务,FRD-CMA模型识别率提高5%,且模型训练时间缩短30%。
Aiming at the problem that a traditional convolutional neural network can be easily overfitted with relatively small data sets,this study presents a fast reduced-dimension convolution model based on attention( FRDCMA),which is suitable for underwater target recognition. The FRD-CMA model describes the attention model of a data set on the basis of the corresponding relationship between convolution kernel and feature map and rapidly reduces the dimensionality to reduce the risk of overfitting when the model is applied to small data sets. The FRDCMA model supports end-to-end processing of underwater target radiated noise. When the acoustic features of radiated noise are extracted and vectorized on the basis of the hydrophone timing relationship,the input features of the model source can be kept intact. The experimental results show that,compared with the previous underwater target recognition task,the FRD-CMA model recognition rate is increased by 5% and the model training time is decreased by 30%.
Aiming at the problem that a traditional convolutional neural network can be easily overfitted with relatively small data sets,this study presents a fast reduced-dimension convolution model based on attention( FRDCMA),which is suitable for underwater target recognition. The FRD-CMA model describes the attention model of a data set on the basis of the corresponding relationship between convolution kernel and feature map and rapidly reduces the dimensionality to reduce the risk of overfitting when the model is applied to small data sets. The FRDCMA model supports end-to-end processing of underwater target radiated noise. When the acoustic features of radiated noise are extracted and vectorized on the basis of the hydrophone timing relationship,the input features of the model source can be kept intact. The experimental results show that,compared with the previous underwater target recognition task,the FRD-CMA model recognition rate is increased by 5% and the model training time is decreased by 30%.
引文
[1]KRIZHEVSKY A,SUTSKEVER I,HINTON G E.ImageN-et classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems.Lake Tahoe,Nevada,2012.
[2]GIRSHICK R,DONAHUE J,DARRELL T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition.Columbus,OH,USA,2014:580-587.
[3]SHELHAMER E,LONG J,DARRELL T.Fully convolutional networks for semantic segmentation[J].IEEE transactions on pattern analysis and machine intelligence,2017,39(4):640-651.
[4]SIMONYAN K,ZISSERMAN A.Very deep convolutional networks for large-scale image recognition[J].arXiv preprint arXiv:1409.1556,2014.
[5]HE Kaiming,ZHANG Xiangyu,REN Shaoqing,et al.Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas,NV,USA,2016.
[6]DENG Jia,DONG Wei,SOCHER R,et al.ImageNet:a large-scale hierarchical image database[C]//Proceedings of 2009 IEEE Conference on Computer Vision and Pattern Recognition.Miami,FL,USA,2009.
[7]LAWRENCE S,GILES C L,TSOI A C,et al.Face recognition:a convolutional neural-network approach[J].IEEEtransactions on neural networks,1997,8(1):98-113.
[8]SZEGEDY C,LIU Wei,JIA Yangqing,et al.Going deeper with convolutions[C]//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition.Boston,MA,USA,2015:1-9.
[9]KAMAL S,MOHAMMED S K,PILLAI P R S,et al.Deep learning architectures for underwater target recognition[C]//Proceedings of 2013 Ocean Electronics.Kochi,India,2013.
[10]FERGUSON E L,RAMAKRISHNAN R,WILLIAMS S B,et al.Convolutional neural networks for passive monitoring of a shallow water environment using a single sensor[C]//Proceedings of 2017 IEEE International Conference on A-coustics,Speech and Signal Processing.New Orleans,LA,USA,2017.
[11]王强,曾向阳.深度学习方法及其在水下目标识别中的应用[C]//中国声学学会水声学分会2015年学术会议论文集.武汉,2015.
[12]CAO Xu,ZHANG Xiaomin,YU Yang,et al.Deep learning-based recognition of underwater target[C]//Proceedings of 2016 IEEE International Conference on Digital Signal Processing.Beijing,China,2016.
[13]NG A,NGIAM J,FOO C Y,et al.UFLDL Tutorial[M/OL].2013[10-04-2013].http://ufldl.stanford.edu/wiki/index.php/UFLDL_Tutorial.
[14]MNIH V,HEESS N,GRAVES A,et al.Recurrent models of visual attention[C]//Proceedings of the 27th International Conference on Neural Information Processing Systems.Montreal,Canada,2014:2204-2212.
[15]LIN M,CHEN Q,YAN S.Network In Network[J/OL]2013,https://ui.adsabs.harvard.edu/abs/2013arXiv1312.4400L.
[2]GIRSHICK R,DONAHUE J,DARRELL T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition.Columbus,OH,USA,2014:580-587.
[3]SHELHAMER E,LONG J,DARRELL T.Fully convolutional networks for semantic segmentation[J].IEEE transactions on pattern analysis and machine intelligence,2017,39(4):640-651.
[4]SIMONYAN K,ZISSERMAN A.Very deep convolutional networks for large-scale image recognition[J].arXiv preprint arXiv:1409.1556,2014.
[5]HE Kaiming,ZHANG Xiangyu,REN Shaoqing,et al.Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas,NV,USA,2016.
[6]DENG Jia,DONG Wei,SOCHER R,et al.ImageNet:a large-scale hierarchical image database[C]//Proceedings of 2009 IEEE Conference on Computer Vision and Pattern Recognition.Miami,FL,USA,2009.
[7]LAWRENCE S,GILES C L,TSOI A C,et al.Face recognition:a convolutional neural-network approach[J].IEEEtransactions on neural networks,1997,8(1):98-113.
[8]SZEGEDY C,LIU Wei,JIA Yangqing,et al.Going deeper with convolutions[C]//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition.Boston,MA,USA,2015:1-9.
[9]KAMAL S,MOHAMMED S K,PILLAI P R S,et al.Deep learning architectures for underwater target recognition[C]//Proceedings of 2013 Ocean Electronics.Kochi,India,2013.
[10]FERGUSON E L,RAMAKRISHNAN R,WILLIAMS S B,et al.Convolutional neural networks for passive monitoring of a shallow water environment using a single sensor[C]//Proceedings of 2017 IEEE International Conference on A-coustics,Speech and Signal Processing.New Orleans,LA,USA,2017.
[11]王强,曾向阳.深度学习方法及其在水下目标识别中的应用[C]//中国声学学会水声学分会2015年学术会议论文集.武汉,2015.
[12]CAO Xu,ZHANG Xiaomin,YU Yang,et al.Deep learning-based recognition of underwater target[C]//Proceedings of 2016 IEEE International Conference on Digital Signal Processing.Beijing,China,2016.
[13]NG A,NGIAM J,FOO C Y,et al.UFLDL Tutorial[M/OL].2013[10-04-2013].http://ufldl.stanford.edu/wiki/index.php/UFLDL_Tutorial.
[14]MNIH V,HEESS N,GRAVES A,et al.Recurrent models of visual attention[C]//Proceedings of the 27th International Conference on Neural Information Processing Systems.Montreal,Canada,2014:2204-2212.
[15]LIN M,CHEN Q,YAN S.Network In Network[J/OL]2013,https://ui.adsabs.harvard.edu/abs/2013arXiv1312.4400L.