1.武汉科技大学 信息科学与工程学院, 湖北 武汉 430081
2.湖南大学 信息科学与工程学院, 湖南 长沙 410082
3.北京航空航天大学 仪器科学与光电工程学院, 北京 100191
[ "刘猛(1995—),男,云南昭通人,硕士研究生,2020年于内蒙古科技大学获得学士学位,主要从事机器学习和深度学习方面的研究。E-mail:1611944656@ qq.com" ]
[ "刘劲(1981—),男,湖南冷水江人,博士,教授,2011年于华中科技大学获得博士学位,主要从事天文导航、图像处理与信号处理方面的研究。E-mail:liujin@ wust.edu.cn" ]
扫 描 看 全 文
刘猛, 刘劲, 尹李君, 等. 基于迭代剪枝VGGNet的火星图像分类[J]. 液晶与显示, 2023,38(4):507-514.
LIU Meng, LIU Jin, YIN Li-jun, et al. Martian image classification based on iterative pruning VGGNet[J]. Chinese Journal of Liquid Crystals and Displays, 2023,38(4):507-514.
刘猛, 刘劲, 尹李君, 等. 基于迭代剪枝VGGNet的火星图像分类[J]. 液晶与显示, 2023,38(4):507-514. DOI: 10.37188/CJLCD.2022-0229.
LIU Meng, LIU Jin, YIN Li-jun, et al. Martian image classification based on iterative pruning VGGNet[J]. Chinese Journal of Liquid Crystals and Displays, 2023,38(4):507-514. DOI: 10.37188/CJLCD.2022-0229.
VGGNet能提供高精度的火星图像分类,但需消耗大量内存资源。鉴于器载计算机内存资源有限,为解决这一矛盾,本文提出了基于迭代剪枝VGGNet的火星图像分类方法。首先,采用迁移学习训练网络的连通性,以便评估神经元的重要性;其次,通过迭代剪枝方法修剪不重要的神经元,以便将全连接层的参数量和内存占用量减少;最后,采用K-means++聚类实现权重参数的量化,利用霍夫曼编码压缩迭代剪枝与量化后的VGGNet权重参数,达到减少存储量和浮点数运算量的作用。此外,通过5种数据增强方法进行数据扩充,目的是解决类别不平衡的问题。实验结果表明,压缩后的VGGNet模型的所占内存、Flops和准确率分别为62.63 Mb、150.6 MFlops和96.15%。与ShuffleNet、MobileNet和EfficientNet等轻量级图像分类算法相比,所提模型具有更好的性能。
VGGNet can provide high-precision Martian image classification, but consumes vast memory resources. Considering the limitation of memory resources of the onboard computer, a Martian image classification method based on iterative pruning VGGNet is proposed to solve this contradiction. Firstly, the transfer learning is used to train the connectivity of the network in order to evaluate the importance of neurons. Secondly, to reduce the number of fully connected layer parameters and memory consumption, the iterative pruning method is used to prune unimportant neurons. Finally, K-means++ clustering is used to quantify the weight parameters, and Huffman coding compresses the weight parameters of VGGNet after iterative pruning and quantization to reduce the storage capacity and floating point arithmetic. Furthermore, the data augmentation is carried out through five data augmentation methods to address the class imbalance. Experimental results show that the memory, Flops and accuracy of the compressed VGGNet model are 62.63 Mb, 150.6 MFlops and 96.15%, respectively. Compared with lightweight image classification algorithms such as ShuffleNet, MobileNet and EfficientNet, the performance of the proposed model is better.
图像分类卷积神经网络迭代方法聚类算法VGGNet
image classificationconvolutional neural networksiterative methodsclustering algorithmsVGGNet
HASKIN L A, WANG A L, JOLLIFF B L, et al. Water alteration of rocks and soils on Mars at the Spirit rover site in Gusev Crater [J]. Nature, 2005, 436(7047): 66-69. doi: 10.1038/nature03640http://dx.doi.org/10.1038/nature03640
SQUYRES S W, KNOLL A H, ARVIDSON R E, et al. Two years at meridiani planum: results from the opportunity rover [J]. Science, 2006, 313(5792): 1403-1407. doi: 10.1126/science.1130890http://dx.doi.org/10.1126/science.1130890
CHRISTENSEN P R, WYATT M B, GLOTCH T D, et al. Mineralogy at meridiani planum from the mini-TES experiment on the opportunity rover [J]. Science, 2004, 306(5702): 1733-1739. doi: 10.1126/science.1104909http://dx.doi.org/10.1126/science.1104909
MCSWEEN JR H Y. Petrology on mars [J]. American Mineralogist, 2015, 100(11/12): 2380-2395. doi: 10.2138/am-2015-5257http://dx.doi.org/10.2138/am-2015-5257
LI J L, ZHANG L, WU Z C, et al. Autonomous Martian rock image classification based on transfer deep learning methods [J]. Earth Science Informatics, 2020, 13(3): 951-963. doi: 10.1007/s12145-019-00433-9http://dx.doi.org/10.1007/s12145-019-00433-9
房建成, 宁晓琳, 刘劲. 航天器自主天文导航原理与方法(第2版)[M]. 北京:国防工业出版社,2017.
FANG J C, NING X L, LIU J. Principles and Methods of Spacecraft Celestial Navigation(Second Edition)[M]. Beijing: National Defense Industry Press, 2017.(in Chinese)
CHAKRAVARTHY A S, ROY R, RAVIRATHINAM P. MRSCAtt: a spatio-channel attention-guided network for Mars rover image classification [C]//Proceedings of 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Nashville: IEEE, 2021: 1961-1970. doi: 10.1109/cvprw53098.2021.00224http://dx.doi.org/10.1109/cvprw53098.2021.00224
SHANG C J, BARNES D. Fuzzy-rough feature selection aided support vector machines for Mars image classification [J]. Computer Vision and Image Understanding, 2013, 117(3): 202-213. doi: 10.1016/j.cviu.2012.12.002http://dx.doi.org/10.1016/j.cviu.2012.12.002
WAGSTAFF K, LU Y, STANBOLI A, et al. Deep Mars: CNN classification of mars imagery for the PDS imaging atlas [J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2018, 32(1). doi: 10.1609/aaai.v32i1.11404http://dx.doi.org/10.1609/aaai.v32i1.11404
LU S, WAGSTAFF K L, ALANIS R, et al. Tutorial: how to access, process, and label PDS image data for machine learning [J]. Machine Learning for Planetary Science, 2022: 91-110. doi: 10.1016/b978-0-12-818721-0.00013-6http://dx.doi.org/10.1016/b978-0-12-818721-0.00013-6
柳思聪, 童小华, 刘世杰, 等. “天问一号” 着陆区遥感形貌建模与制图分析[J]. 深空探测学报, 2022, 9(3): 1-10.
LIU S C, TONG X H, LIU S J, et al. Topography modeling, mapping and analysis of china’s first mars mission tianwen-1 landing area from remote sensing images[J]. Journal of Deep Space Exploration, 2022, 9(3): 338-347.(in Chinese)
王旖旎. 基于Inception V3的图像状态分类技术[J]. 液晶与显示, 2020, 35(4):389-394. doi: 10.3788/yjyxs20203504.0389http://dx.doi.org/10.3788/yjyxs20203504.0389
WANG Y N. Image classification technology based on inception V3[J]. Chinese Journal of Liquid Crystals and Displays, 2020, 35(4):389-394. (in Chinese). doi: 10.3788/yjyxs20203504.0389http://dx.doi.org/10.3788/yjyxs20203504.0389
SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition [J/OL]. arXiv, 2014: 1409.1556.
MA N N, ZHANG X Y, ZHENG H T, et al. ShuffleNet V2: practical guidelines for efficient CNN architecture design [C]//Proceedings of the 15th European Conference on Computer Vision. Munich: Springer, 2018: 122-138. doi: 10.1007/978-3-030-01264-9_8http://dx.doi.org/10.1007/978-3-030-01264-9_8
KOONCE B. MobileNetV3 [M]//KOONCE B. Convolutional Neural Networks with Swift for Tensorflow. Berkeley, CA: Apress, 2021: 125-144. doi: 10.1007/978-1-4842-6168-2_11http://dx.doi.org/10.1007/978-1-4842-6168-2_11
KOONCE B. EfficientNet [M]//KOONCE B. Convolutional Neural Networks with Swift for Tensorflow. Berkeley, CA: Apress, 2021: 109-123. doi: 10.1007/978-1-4842-6168-2_10http://dx.doi.org/10.1007/978-1-4842-6168-2_10
MCSWEEN H Y, ARVIDSON R E, BELL III J F, et al. Basaltic rocks analyzed by the spirit rover in Gusev Crater [J]. Science, 2004, 305(5685): 842-845. doi: 10.1126/science.3050842http://dx.doi.org/10.1126/science.3050842
MCSWEEN H Y, WYATT M B, GELLERT R, et al. Characterization and petrologic interpretation of olivine-rich basalts at Gusev Crater, Mars [J]. Journal of Geophysical Research: Planets, 2006, 111(E2): E02S10. doi: 10.1029/2005je002477http://dx.doi.org/10.1029/2005je002477
MCSWEEN H Y, RUFF S W, MORRIS R V, et al. Alkaline volcanic rocks from the Columbia Hills, Gusev Crater, Mars [J]. Journal of Geophysical Research: Planets, 2006, 111(E9): E09S91. doi: 10.1029/2006je002698http://dx.doi.org/10.1029/2006je002698
MCSWEEN H Y, RUFF S W, MORRIS R V, et al. Mineralogy of volcanic rocks in Gusev Crater, Mars: reconciling mössbauer, alpha particle X-ray spectrometer, and miniature thermal emission spectrometer spectra [J]. Journal of Geophysical Research: Planets, 2008, 113(E6): E06S04. doi: 10.1029/2007je002970http://dx.doi.org/10.1029/2007je002970
SQUYRES S W, AHARONSON O, CLARK B C, et al. Pyroclastic activity at home plate in Gusev Crater, Mars [J]. Science, 2007, 316(5825): 738-742. doi: 10.1126/science.1139045http://dx.doi.org/10.1126/science.1139045
SCHMIDT M E, CAMPBELL J L, GELLERT R, et al. Geochemical diversity in first rocks examined by the Curiosity Rover in Gale Crater: Evidence for and significance of an alkali and volatile-rich igneous source [J]. Journal of Geophysical Research: Planets, 2014, 119(1): 64-81. doi: 10.1002/2013je004481http://dx.doi.org/10.1002/2013je004481
LI B H, HOU Y T, CHE W X. Data augmentation approaches in natural language processing: a survey [J]. AI Open, 2022, 3: 71-90. doi: 10.1016/j.aiopen.2022.03.001http://dx.doi.org/10.1016/j.aiopen.2022.03.001
ZUO C, QIAN J M, FENG S J, et al. Deep learning in optical metrology: a review [J]. Light: Science & Applications, 2022, 11(1): 39. doi: 10.1038/s41377-022-00714-xhttp://dx.doi.org/10.1038/s41377-022-00714-x
HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition [C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770-778. doi: 10.1109/cvpr.2016.90http://dx.doi.org/10.1109/cvpr.2016.90
IANDOLA F, MOSKEWICZ M, KARAYEV S, et al. DenseNet: implementing efficient ConvNet descriptor pyramids [J/OL]. arXiv, 2014: 1404.1869.
王苹. 高精度视频配准算法中的静态图像配准算法[J]. 液晶与显示, 2020, 35(6):612-618. doi: 10.3788/yjyxs20203506.0612http://dx.doi.org/10.3788/yjyxs20203506.0612
WANG P. Static image registration algorithm in high-precision video registration algorithm[J]. Chinese Journal of Liquid Crystals and Displays, 2020, 35(6):612-618. (in Chinese). doi: 10.3788/yjyxs20203506.0612http://dx.doi.org/10.3788/yjyxs20203506.0612
ONGKADINATA D, PUTRI F P. Quality and size assessment of quantized images using K-Means++ clustering [J]. Bulletin of Electrical Engineering and Informatics, 2020, 9(3): 1183-1188. doi: 10.11591/eei.v9i3.1985http://dx.doi.org/10.11591/eei.v9i3.1985
0
浏览量
233
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构