Page 101 - 《应用声学》2025年第2期

第 44 卷 第 2 期 申小虎等： 网络剪枝与知识蒸馏相结合的轻量级鸟声识别方法 361

2019. Processing. New Orleans, USA, 2017: 4820–4824.
[17] Tan M, Chen B, Pang R, et al. MnasNet: Platform- [25] Aflalo Y, Noy A, Lin M, et al. Knapsack pruning with
aware neural architecture search for mobile[C]// 2019 inner distillation[J]. arXiv: 2002.08258, 2020.
IEEE/CVF Conference on Computer Vision and Pat- [26] Zhu J, Zhao Y, Pei J. Progressive kernel pruning based
tern Recognition (CVPR). Long Beach, CA, USA, 2019: on the information mapping sparse index for CNN com-
2815–2823. pression[J]. IEEE Access, 2021, 9: 10974–10987.
[18] Hu J, Shen L, Albanie S, et al. Squeeze-and-excitation [27] Li T, Li J, Liu Z, et al. Few sample knowledge distillation
networks[J]. IEEE Transactions on Pattern Analysis and for efficient network compression[C]// Proceedings of the
Machine Intelligence, 2019, 42(8): 2011–2023.
IEEE/CVF Conference on Computer Vision and Pattern
[19] 杨宏炳, 迟勇欣, 王金光. 基于剪枝网络的知识蒸馏对遥
Recognition. Seattle, USA, 2020: 14639–14647.
感卫星图像分类方法 [J]. 计算机应用研究, 2021, 38(8):
[28] Kahl S, Stter F R, Goau H, et al. Overview of BirdCLEF
2469–2473.
2019: Large-scale bird recognition in soundscapes[C]//
Yang Hongbing, Chi Yongxin, Wang Jinguang. Knowl-
Conference and Labs of the Evaluation Forum, 2019.
edge distillation method for remote sensing satellite im-
[29] Molchanov P, Tyree S, Karras T, et al. Pruning con-
age classification based on pruning network[J]. Applica-
volutional neural networks for resource efficient transfer
tion Research of Computers, 2021, 38(8): 2469–2473.
learning[J]. arXiv: 1611.06440, 2017.
[20] 姜晓勇, 李忠义, 黄朗月, 等. 神经网络剪枝技术研究综述 [J].
[30] Lasseck M. Bird species identification in soundscapes[C]//
应用科学学报, 2022, 40(5): 838–849.
Conference and Labs of the Evaluation Forum, 2019.
Jiang Xiaoyong, Li Zhongyi, Huang Langyue, et al. Re-
[31] Iandola F N, Han S, Moskewicz M W, et al. SqueezeNet:
view of neural network pruning techniques[J]. Journal of
Alexnet-level accuracy with 50x fewer parameters and
Applied Sciences, 2022, 40(5): 838–849.
< 0.5 MB model size[J]. arXiv: 1602.07360, 2016.
[21] 黄震华, 杨顺志, 林威, 等. 知识蒸馏研究综述 [J]. 计算机学
[32] Han K, Wang Y, Zhang Q, et al. Model rubik’s cube:
报, 2022, 45(3): 624–653.
Huang Zhenhua, Yang Shunzhi, Lin Wei, et al. Knowledge Twisting resolution, depth and width for TinyNets[J].
distillation: A survey[J]. Chinese Journal of Computers, arXiv: 2010.14819, 2020.
2022, 45(3): 624–653. [33] Howard A G, Zhu M, Chen B, et al. MobileNets: Efficient
[22] Hinton G, Vinyals O, Dean J. Distilling the knowledge in a convolutional neural networks for mobile vision applica-
neural network[J]. Computer Science, 2015, 14(7): 38–39. tions[J]. arXiv: 1704.04861, 2017.
[23] Zhang L, Song J, Gao A, et al. Be your own teacher: Im- [34] Sandler M, Howard A G, Zhu M, et al. Mo-
prove the performance of convolutional neural networks bilenetv2: Inverted residuals and linear bottlenecks[C]//
via self distillation[C]// 2019 IEEE/CVF International 2018 IEEE/CVF Conference on Computer Vision and
Conference on Computer Vision (ICCV), Seoul, KR, 2019: Pattern Recognition, Salt Lake City, UT, USA, 2018:
3712–3721. 4510–4520.
[24] Lu L, Guo M, Renals S. Knowledge distillation for small- [35] Ma N N, Zhang X Y, Zheng H T, et al. Shufflenet
footprint highway networks[C]// Proceedings of the IEEE v2: Practical guidelines for efficient CNN architecture de-
International Conference on Acoustics, Speech and Signal sign[J]. arXiv: 1807.11164, 2018.