计算机科学 ›› 2021, Vol. 48 ›› Issue (7): 190-198.doi: 10.11896/jsjkx.200800225

• 计算机图形学&多媒体 • 上一篇    下一篇

基于胶囊网络及其权重剪枝的SAR图像变化检测方法

陈志文1, 王坤1, 周广蕴2, 王旭2, 张晓丹2, 朱虎明1   

  1. 1 西安电子科技大学智能感知与图像理解教育部重点实验室 西安710071
    2 北京机电工程研究所 北京100074
  • 收稿日期:2020-08-31 修回日期:2021-01-24 出版日期:2021-07-15 发布日期:2021-07-02
  • 通讯作者: 朱虎明(zhuhum@mail.xidian.edu.cn)
  • 基金资助:
    教育部产学合作协同育人项目(201901159006);国防基础科研计划资助(JCKY2016204A102)

SAR Image Change Detection Method Based on Capsule Network with Weight Pruning

CHEN Zhi-wen1, WANG Kun1, ZHOU Guang-yun2, WANG Xu2, ZHANG Xiao-dan2, ZHU Hu-ming1   

  1. 1 Key Laboratory of Intelligent Perception and Image Understanding of Ministry of Education,Xidian University,Xi’an 710071,China
    2 Beijing Electro-mechanical Engineering Institute,Beijing 100074,China
  • Received:2020-08-31 Revised:2021-01-24 Online:2021-07-15 Published:2021-07-02
  • About author:CHEN Zhi-wen,born in 1995,postgra-duate.His main research interests include artificial intelligence computing system and its performance optimization.(18829535531@163.com)
    ZHU Hu-ming,born in 1978,Ph.D,associate professor.His main research interests include high performance computing and parallel machine learning algorithm.
  • Supported by:
    Cooperative Education Project of the Ministry of Education(201901159006) and National Defense Basic Scientific Research Program Funding(JCKY2016204A102).

摘要: 基于深度神经网络的SAR图像变化检测算法由于精确率高等优点,已被广泛应用在农业检测、城市规划以及森林预警等多个领域。设计了基于胶囊网络的SAR图像变化检测算法,针对其模型复杂度高、参数量大等问题,提出了基于权重剪枝的模型压缩方法。该方法对其胶囊网络参数进行逐层分析,针对不同类型的层采取不同的剪枝策略,对网络中冗余的参数进行剪枝,随后对剪枝后的网络进行微调,从而提高了剪枝后模型的检测性能。最后,通过对模型中保留下来的参数进行压缩存储,显著降低了模型所占用的存储空间。在4组真实SAR图像上的实验结果证明了所提出的模型压缩方法的有效性。

关键词: SAR图像, 变化检测, 胶囊网络, 模型压缩, 权重剪枝

Abstract: The SAR image change detection algorithm based on deep neural network has been widely used in many fields such as agricultural detection,urban planning and forest early warning due to its high accuracy.This paper designs a SAR image change detection algorithm based on capsule network.In view of its high model complexity and large number of parameters,a model compression method based on weight pruning is proposed.This method performs layer-by-layer analysis of its capsule network parameters,adopts different pruning strategies for different types of layers,prunes redundant parameters in the network,and then fine-tunes the pruned network to improve the detection performance of the model.Finally,by compressing and storing the para-meters retained in the model,the storage space occupied by the model is significantly reduced.Experiments on four datasets of real SAR images prove the effectiveness of the proposed model compression method.

Key words: Capsule network, Change detection, Model compression, SAR image, Weight pruning

中图分类号: 

  • TP391
[1]LIU L,JIA Z,YANG J,et al.SAR Image Change DetectionBased on Mathematical Morphology and the K-Means Clustering Algorithm[J].IEEE Access,2019,7:43970-43978.
[2]SABOUR S,FROSST N,HINTON G E.Dynamic routing between capsules[C]//Advances in Neural Information Processing Systems.2017:3856-3866.
[3]DHILLON A,VERMA G K.Convolutional neural network:areview of models,methodologies and applications to object detection[J].Progress in Artificial Intelligence,2020,9(2):85-112.
[4]PAOLETTI M E,HAUT J M,FERNANDEZ-BELTRAN R,et al.Capsule Networks for Hyperspectral Image Classification[J].IEEE Transactions on Geoscience & Remote Sensing,2019,57(4):2145-2160.
[5]CHOI Y,EL-KHAMY M,LEE J.Universal Deep Neural Network Compression[J].IEEE Journal on Selected Topics in Signal Processing,2020,14(4):715-726.
[6]HOWARD A G,ZHU M,CHEN B ,et al.Mobilenets:Efficient convolutional neural networks for mobile vision applications[J].arXiv:1704.04861,2017.
[7]ZHANG T Y,YE S K,ZHANG K Q,et al.A Systematic DNN Weight Pruning Framework using Alternating Direction Method of Multipliers[J].arXiv:1804.03294,2018.
[8]GUPTA S,AGRAWAL A,GOPALAKRISHNAN K,et al.Deep learning with limited numerical precision[C]//Internatio-nal Conference on Machine Learning.2015:1737-1746.
[9]SUN S,CHENG Y,GAN Z,et al.Patient Knowledge Distillation for BERT Model Compression[J].arXiv:1908.09355,2019.
[10]HUANG G,LIU Z,WEINBERGER K Q.Densely Connected Convolutional Networks[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR).2017:2261-2269.
[11]HU J,SHEN L,ALBANIE S,et al.Squeeze-and-Excitation Networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(8):2011-2023.
[12]XIONG Y,SU G,YE S,et al.Deeper Capsule Network forComplex Data[C]//2019 International Joint Conference on Neural Networks (IJCNN).Budapest,Hungary,2019:1-8.
[13]KRIZHEVSKY A,SUTSKEVER I,HINTON G E.Imagenetclassification with deep convolutional neural networks[C]//Advances in Neural Information Processing Systems.2012:1097-1105.
[14]KINGMA D P,BA J.Adam:A method for stochastic optimization[J].arXiv:1412.6980,2014.
[15]KRINIDIS S,CHATZIS V.A robust fuzzy local informationC-means clustering algorithm[J].IEEE Transactions on Image Processing,2010,19(5):1328-1337.
[16]HE Y,KANG G,DONG X,et al.Soft Filter Pruning for Acce-lerating Deep Convolutional Neural Networks[J].arXiv:1808.06866.
[17]PASANDI M M,HAJABDOLLAHI M,KARIMI N,et al.Mode-ling of Pruning Techniques for Deep Neural Networks Simplification[J].arXiv:2001.04062.
[18]VU V T,GOMES N R,PETTERSSON M I,et al.Bivariate Gamma Distribution for Wavelength-Resolution SAR Change Detection[J].IEEE Transactions on Geoscience and Remote Sensing,2019,57(1):473-481.
[1] 楚玉春, 龚航, 王学芳, 刘培顺.
基于YOLOv4的目标检测知识蒸馏算法研究
Study on Knowledge Distillation of Target Detection Algorithm Based on YOLOv4
计算机科学, 2022, 49(6A): 337-344. https://doi.org/10.11896/jsjkx.210600204
[2] 蓝凌翔, 池明旻.
基于特征注意力融合网络的遥感变化检测研究
Remote Sensing Change Detection Based on Feature Fusion and Attention Network
计算机科学, 2022, 49(6): 193-198. https://doi.org/10.11896/jsjkx.210500058
[3] 武霖, 孙静宇.
多分支RA胶囊网络及在图像分类中的应用
Multi-branch RA Capsule Network and Its Application in Image Classification
计算机科学, 2022, 49(6): 224-230. https://doi.org/10.11896/jsjkx.210400087
[4] 程祥鸣, 邓春华.
基于无标签知识蒸馏的人脸识别模型的压缩算法
Compression Algorithm of Face Recognition Model Based on Unlabeled Knowledge Distillation
计算机科学, 2022, 49(6): 245-253. https://doi.org/10.11896/jsjkx.210400023
[5] 范红杰, 李雪冬, 叶松涛.
面向电子病历语义解析的疾病辅助诊断方法
Aided Disease Diagnosis Method for EMR Semantic Analysis
计算机科学, 2022, 49(1): 153-158. https://doi.org/10.11896/jsjkx.201100125
[6] 詹瑞, 雷印杰, 陈训敏, 叶书函.
基于多重差异特征网络的街景变化检测
Street Scene Change Detection Based on Multiple Difference Features Network
计算机科学, 2021, 48(2): 142-147. https://doi.org/10.11896/jsjkx.200500158
[7] 刘海潮, 王莉.
基于深度图卷积胶囊网络的图分类模型
Graph Classification Model Based on Capsule Deep Graph Convolutional Neural Network
计算机科学, 2020, 47(9): 219-225. https://doi.org/10.11896/jsjkx.190900044
[8] 孙彦丽, 叶炯耀.
基于剪枝与量化的卷积神经网络压缩方法
Convolutional Neural Networks Compression Based on Pruning and Quantization
计算机科学, 2020, 47(8): 261-266. https://doi.org/10.11896/jsjkx.190700062
[9] 周志一, 宋冰, 段鹏松, 曹仰杰.
基于WiFi信号的轻量级步态识别模型LWID
LWID:Lightweight Gait Recognition Model Based on WiFi Signals
计算机科学, 2020, 47(11): 25-31. https://doi.org/10.11896/jsjkx.200200044
[10] 周岳勇,程江华,刘通,王洋,陈明辉.
高分辨率SAR图像道路提取综述
Review of Road Extraction for High-resolution SAR Images
计算机科学, 2020, 47(1): 124-135. https://doi.org/10.11896/jsjkx.190100033
[11] 李青华, 李翠平, 张静, 陈红, 王绍卿.
深度神经网络压缩综述
Survey of Compressed Deep Neural Network
计算机科学, 2019, 46(9): 1-14. https://doi.org/10.11896/j.issn.1002-137X.2019.09.001
[12] 曾燕, 陈岳林, 蔡晓东.
一种基于权重哈希化的深度人脸识别算法
Deep Face Recognition Algorithm Based on Weighted Hashing
计算机科学, 2019, 46(6): 277-281. https://doi.org/10.11896/j.issn.1002-137X.2019.06.041
[13] 张天柱, 邹承明.
使用模糊聚类的胶囊网络在图像分类上的研究
Study on Image Classification of Capsule Network Using Fuzzy Clustering
计算机科学, 2019, 46(12): 279-285. https://doi.org/10.11896/jsjkx.190200315
[14] 郭杨,秦品乐.
基于多模态局部转向核的脑部多发性硬化检测算法研究
Change Detection of Multiple Sclerosis in Brain Based on Multi-modal Local Steering Kernel
计算机科学, 2018, 45(3): 241-246. https://doi.org/10.11896/j.issn.1002-137X.2018.03.038
[15] 赵杰,王配配,门国尊.
基于非局部相似和低秩矩阵逼近的SAR图像去噪
SAR Image Denosing Based on Nonlocal Similarity and Low Rank Matrix Approximation
计算机科学, 2017, 44(Z6): 183-187. https://doi.org/10.11896/j.issn.1002-137X.2017.6A.042
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!