Application of an Improved ELU Convolution Neural Network in the SAR Image Ship Detection

doi:10.13474/j.cnki.11-2246.2018.0024

Abstract

Abstract: With the development of space technology,the SAR load detection system in our country is showing the development trend of variousness and multi resolution.Since the traditional detection identification methods are difficult to satisfy the multiresolution and various characteristics of SAR image data,it's necessary to seek a different method to extract effective features from the multiresolution image data.For the rapidly development of intelligence,our scheme bases on the characteristics of SAR image proposes the convolution of the improved neural network method and utilizes ELU as an activation function to establish the deep learning model,which combined with the ELU and quadratic cost function.At the same time,based on the training sample, we establish the sample characteristics and centric distance functions,and then use the fuzzy support vector machine to classify the extracted characteristics.The experimental results show that the proposed method can improve the noise resistance of SAR image for the ship detection, and the detection rate can reach up to 98.6%.

Key words: synthetic aperture radar, the convolution neural network, fuzzy support vector machine, cost function, classification function

CLC Number:

P237

BAI Yu, JIANG Dongmin, PEI Jiajun, ZHANG Ning, BAI Yu. Application of an Improved ELU Convolution Neural Network in the SAR Image Ship Detection[J]. 测绘通报, 2018, 0(1): 125-128.

References

[1] DEKKER R J.Texture Analysis and Classification of ERS SAR Images for Map Updating of Urban Areas in the Netherlands[J].IEEE Transactions on Geoscience and Remote Sensing,2003,41(9):1950-1958.
[2] MA Y,HE J,WU L,et al.An Effective Face Verification Algorithm to Fuse Complete Features in Convolution Neural Network[C]//Multi Media Modeling.Cham:Springer International Puishing,2016:39-46.
[3] LECUN Y,BOTTOU L,BENGIO Y,et al.Gradient-based Learning Applied to Document Recognition[J].Proceedings of the IEEE,1998,86(11):2278-2324.
[4] 尹宝才,王文通,王立春.深度学习研究综述[J].北京工业大学学报,2015,41(1):48-59.
[5] CHU W,ONG C J,KEERTHY S S.An Improved Conjugate Gradient Method Scheme to the Solution of Least Squares SVM[J].IEEE Transactions on Neural Networks,2005,16(2):498-501.
[6] KEERTHI S S,SHEVADE S K.SMO for Least Squares SVM Formulations[J].Neural Computation,2003,15(2):487-507.
[7] RUMELHART D E,HINTO G E,WILLIAMS R J,et al.Learning Representations by Back-propagating Errors[J].Nature,1986,323(6088):533-536.
[8] CORTES C,VAPNIK V.Support Vector Network[J].Machine Learning,1995,20(3):273-297.
[9] KAPLAN L M.Analysis of Multiplicative Speckle Models for Template-based SAR AT[J].IEEE Transactions on Aerospace and Electronic Systems 2001,37(4):1424-1432.
[10] WESTON J,RATLE F,MOBAHI H,et al.Deep Learning via Semi-supervised Embedding[C]//International Conference on Machine Learning.[S.l.]:ACM,2012:1168-1175.
[11] GLOROT X,BENGIO Y.Understanding the Difficulty of Training Deep Feed Forward Neural Networks[J].Journal of Machine Learning Research,2010(9):249-256.
[12] WU Tao,CHEN Xi,RUANG Xiangwei,et al.Study on SAR Target Recognition Based on Support Vector Machine[C]//20092nd Asian-Pacific Conference Synthetic Aperture Radar.[S.l.]:IEEE,2009:856-859.
[13] 王璐,张帆,李伟,等.基于Gabor滤波器和局部纹理特征提取的SAR目标识别算法[J].雷达学报,2015,4(6):658-665.
[14] BENGIO Y.Practical Recommendations for Gradient-Based Training of Deep Architectures[J].Lecture Notes in Computer Science,2012,7700(1-3):437-478.
[15] 齐会娇,王英华,丁军,等.基于多信息字典学习及稀疏表示的SAR目标识别[J].系统工程与电子技术,2015,37(6):1280-1287.
[16] DING J,LIU H W,WANG Y H.SAR Image Target Recognition Based on Non-negative Sparse Representation[J].Journal of Electronics & Information Technology,2014,36(9):2194-2200.
[17] TISON C,NICOLAS J M,TUPIN F,et al.A New Statistical Model for Markovian Classification of Urban Areas in high Resolution SAR Images[J].IEEE Trans Geo science and Remote Sensing 2004,42(10):2046-2057.