基于自适应层信息熵的卷积神经网络压缩

doi:10.12263/DZXB.20201372

摘要/Abstract

摘要：

网络剪枝是一种有效的卷积神经网络压缩方法.多数现有压缩方法因迭代剪枝了“不重要”的网络结构，一方面破坏了网络结构的信息整体性，另一方面其迭代操作耗费了大量的计算资源与时间.为了解决上述问题，论文从网络结构全局考虑，提出基于自适应层信息熵的卷积神经网络压缩方法.首先，在获取压缩网络结构的过程中，本文设计了一种端到端的结构化网络剪枝方案，将卷积层看作一个整体，利用层信息熵之间的关联性直接确定各卷积层过滤器的保留率，避免迭代剪枝训练造成的信息损失.其次，对剪裁后的网络进行重训练时，综合考虑压缩过程中使用的层信息熵指标，通过对卷积层与批归一化（Batch Normalization，BN）层进行自适应联合嫁接，让网络学习到更多的信息，提升网络性能.针对3种主流网络在不同的数据集上进行了实验，验证了所提方法的有效性与优越性.例如在CIFAR-10上，针对ResNet-56，相比于基线网络，在计算量压缩36.2%的情况下，本文方法准确率提升了1%；针对ResNet-110，在计算量压缩52.4%的情况下，本文方法准确率提升了1.42%；针对轻量型网络MobileNetV2，在计算量压缩55.2%的情况下，本文方法准确率提升了1.29%.

关键词: 卷积神经网络, 网络剪枝, 信息熵, 嫁接, 模型压缩

Abstract:

Network pruning has proven to be an effective approach to compress convolutional neural network (CNN). However, most existing CNN compression methods iteratively prune the "least important" filters, which not only destroys the information integrity of network structures, but also results in significant computation cost due to the iterative operation. To solve the problems, a convolutional neural network compression method based on adaptive layer entropy(ALE) is proposed, considering a global network structure. Firstly, an end-to-end network pruning strategy is designed, in which the retention rate of each convolutional layer filter is directly determined based on the entropy correlation between layers. The pruning strategy takes the convolutional layer as a whole, which decreases the information loss and computation cost of iterative pruning. Then, an adaptive joint grafting method, in which both convolutional and batch normalization(BN) layers are considered, is presented for the pruned network retraining to learn more information from the network. The layer entropies used in the compression are also utilized for the grafting. Experiments are conducted on different benchmarks and three popular networks, which demonstrate the efficiency and superiority of the proposed ALE over other methods. For the experiments on CIFAR-10, ALE achieves 36.2%, 52.4% and 55.2% pruned rate in FLOPs for ResNet-56, ResNet-110 and MobileNetV2 while with increase of 1%, 1.42%, 1.29% accuracy respectively.

Key words: convolutional neural network, network pruning, entropy, grafting, model compression

中图分类号:

TP391.41

魏钰轩, 陈莹. 基于自适应层信息熵的卷积神经网络压缩[J]. 电子学报, 2022, 50(10): 2398-2408.

Yu-xuan WEI, Ying CHEN . Convolutional Neural Network Compression Based on Adaptive Layer Entropy[J]. Acta Electronica Sinica, 2022, 50(10): 2398-2408.

图/表 26

图1 逐层迭代剪枝过程

图2 端到端迭代剪枝过程

图3 ALE总体流程框架

表1 ALE中符号定义及描述

符号	描述	符号	描述
N	卷积神经网络模型	$O H S$	模型原始层信息熵区间
L	卷积层个数	$α m a x$	卷积层过滤器最大保留率
$c i$	第 $i$ 层过滤器个数	$H S$	模型层信息熵区间
$W i$	第 $i$ 层网络的权重	$α i$	第i层过滤器保留率
$H ∙$	信息熵	$N i_l 1$	网络 $N i$ 的卷积第一层
$β$	自适应嫁接系数	$N i_b 1$	网络 $N i$ 的BN第一层
SE	最小信息熵	$p k i$	$W i$ 落在第k份区间的概率
BE	最大信息熵	$c i,$	剪枝后第 $i$ 层过滤器的个数

表1 ALE中符号定义及描述

符号	描述	符号	描述
N	卷积神经网络模型	$O H S$	模型原始层信息熵区间
L	卷积层个数	$α m a x$	卷积层过滤器最大保留率
$c i$	第 $i$ 层过滤器个数	$H S$	模型层信息熵区间
$W i$	第 $i$ 层网络的权重	$α i$	第i层过滤器保留率
$H ∙$	信息熵	$N i_l 1$	网络 $N i$ 的卷积第一层
$β$	自适应嫁接系数	$N i_b 1$	网络 $N i$ 的BN第一层
SE	最小信息熵	$p k i$	$W i$ 落在第k份区间的概率
BE	最大信息熵	$c i,$	剪枝后第 $i$ 层过滤器的个数

图4 ResNet-56在CIFAR-100上层信息熵变化过程

图5 网络各层过滤器保留率获取过程

图6 压缩网络结构的过程

图7 具有短连接块的网络压缩方法

图8 联合嫁接示意图

图9 自适应嫁接系数β变化关系图

图10 多网络联合嫁接示意图

表2 ResNet-56在CIFAR-10上结果

模型	准确率	计算量(压缩率)	参数量(压缩率)
ResNet-56	93.26%	125.49M(0.0%)	0.85M(0.0%)
PF-A^[16]	93.10%	112.00M(10.7%)	0.77M(9.4%)
PF-B^[16]	93.06%	90.90M(27.6%)	0.73M(14.1%)
SSS^[32]	93.39%	89.35M(28.8%)	0.59M(30.6%)
HRank1^[33]	93.52%	88.72M(29.3%)	0.71M(16.8%)
LEGR1^[34]	94.10%	87.80M(30.0%)
NISP^[35]	93.01%	81.00M(35.5%)	0.49M(42.4%)
ALE-100%	94.26%	80.10M(36.2%)	0.54M(32.9%)
GAL-0.6^[21]	93.38%	78.30M(37.6%)	0.75M(11.8%)
HRank2^[33]	93.17%	62.72M(50.0%)	0.49M(42.4%)
Hinge^[36]	93.69%	62.72M(50.0%)	0.44M(51.27%)
KSE^[26]	93.23%	60.00M(50.0%)	0.43M(49.4%)
FPGM^[19]	93.26%	59.40M(52.6%)
LEGR2^[34]	93.70%	58.90M(53.1%)
ABC-70%^[22]	93.23%	58.54M(54.1%)	0.39M(54.2%)
ALE-60%	93.64%	49.53M(60.5%)	0.36M(57.6%)
C-SGD^[4]	93.31%	49.13M(60.85%)

表3 ResNet-110在CIFAR-10上结果

模型	准确率	计算量(压缩率)	参数量(压缩率)
ResNet-110	93.50%	252.89M(0.0%)	1.72M(0.0%)
GAL-0.1^[21]	93.59%	205.7M(18.7%)	1.65M(4.07%)
NISP^[35]	93.32%	143.35M(43.3%)
GAL-0.5^[21]	92.74%	130.20M(48.5%)	0.95M(44.8%)
FPGM^[19]	93.74%	121.00M(52.2%)
ALE-80%	94.92%	120.29M(52.4%)	0.89M(48.3%)
HRank^[33]	93.36%	105.70M(58.2%)	0.70M(59.3%)
ALE-60%	94.36%	92.68M(63.4%)	0.69M(59.9%)
ABC-60%^[22]	93.58%	89.87M(64.5%)	0.56M(67.4%)

表4 VGGNet在CIFAR-10上结果

模型	准确率	计算量(压缩率)	参数量(压缩率)
VGGNet	93.96%	313.74M(0.0%)	14.95M(0.0%)
SSS^[32]	93.02%	183.13M(42.3%)	3.93M(73.7%)
GAL-0.1^[21]	93.42%	171.89M(45.2%)	2.67M(82.2%)
HRank1^[33]	92.34%	108.61M(65.4%)	2.64M(82.3%)
HRank2^[33]	91.23%	73.70M(76.5%)	1.78M(88.2%)
ALE-90%	92.85%	71.29M(77.28%)	2.87M(80.8%)
LEGR^[34]	92.40%	70.30M(77.6%)	---
GAL-0.2^[21]	91.89%	65.85M(79.0%)	3.35M(77.6%)
ALE-80%	92.81%	60.23M(80.88%)	2.46M(83.6%)

表5 MobileNetV2在CIFAR-10上结果

模型	准确率	计算量(压缩率)	参数量(压缩率)
MobileNetV2	92.25%	91.15M(0.0%)	2.26M(0.0%)
ALE-90%	93.54%	40.86M(55.2%)	0.94M(58.4%)
ALE-70%	92.79%	32.66M(64.2%)	0.83M(63.3%)

表6 ResNet-56在CIFAR-100上结果

模型	准确率	计算量(压缩率)	参数量(压缩率)
ResNet-56	71.92%	125.49M(0.0%)	0.85M(0.0%)
PF-A^[16]	70.42%	112.44M(10.4%)	0.77M(9.4%)
PF-B^[16]	69.95%	90.85M(27.6%)	0.73M(13.7%)
ALE-90%	70.91%	64.77M(48.4%)	0.53M(37.7%)
ALE-80%	70.72%	57.97M(53.8%)	0.47M(44.7%)

表7 ResNet-56和ResNet-110在SVHN上结果

模型	准确率	计算量(压缩率)	参数量(压缩率)
ResNet-56	96.38%	125.49M(0.0%)	0.85M(0.0%)
GAL-0.6^[21]	95.82%	73.58M(41.4%)	0.56M(34.1%)
ALE-60%	96.52%	50.17M(60.0%)	0.35M(58.8%)
ResNet-110	96.36%	252.89M(0.0%)	1.72M(0.0%)
GAL-0.06^[21]	96.63%	120.77M(52.2%)	0.83M(51.7%)
ALE-60%	96.65%	98.50M(61.1%)	0.71M(58.7%)

表8 ResNet-50在Market1501上结果

模型

mAP

Rank1

计算量

(压缩率)

参数量

(压缩率)

ResNet-50

65.1%

84.1%

4087.14M

(0.0%)

24.99M

(0.0%)

ALE-90%

62.2%

82.1%

1577.95M

(61.4%)

8.28M

(66.9%)

ALE-80%

61.5%

81.7%

1419.49M

(65.3%)

7.48M

(70.1%)

表9 ResNet-50在Duke上结果

模型

mAP

Rank1

计算量

(压缩率)

参数量

(压缩率)

ResNet-50

52.3%

72.1%

4087.14M

(0.0%)

24.89M

(0.0%)

ALE-90%

51.7%

71.1%

1639.11M

(59.9%)

7.41M

(70.2%)

ALE-80%

51.0%

70.5%

1508.50M

(63.1%)

7.22M

(71.0%)

图11 ResNet-56在CIFAR-10上最大保留率αmax对网络性能的影响

图12 MobileNetV2在CIFAR-10上压缩前后层信息熵关系比较

图13 嫁接对比实验结果

表10 ResNet-56在CIFAR-100上结果

模型	准确率	计算量(压缩率)	参数量(压缩率)
ResNet-56	71.92%	125.49M(0.0%)	0.85M(0.0%)
ALE-80%-10	69.31%	69.93M(44.3%)	0.52M(38.8%)
ALE-80%-15	70.06%	68.31M(45.6%)	0.53M(37.7%)
ALE-90%-122	70.91%	64.77M(48.4%)	0.53M(37.7%)
ALE-80%-60	70.48%	63.74M(49.2%)	0.48M(43.5%)
ALE-80%-20	70.26%	61.97M(50.6%)	0.50M(41.2%)
ALE-80%-90	70.93%	61.67M(50.9%)	0.49M(42.4%)
ALE-80%-30	70.19%	61.50M(51.0%)	0.47M(44.7%)
ALE-80%-122	70.72%	57.97M(53.8%)	0.47M(44.7%)
ALE-80%-150	70.65%	55.35M(55.9%)	0.44M(48.2%)
ALE-80%-5	69.17%	50.32M(59.9%)	0.36M(57.6%)

表11 层最小保留率的影响

模型	准确率	计算量(压缩率)	参数量(压缩率)
ResNet-56	93.26%	125.49M(0.0%)	0.85M(0.0%)
ALE-60%	93.64%	49.53M(60.5%)	0.36M(57.6%)
20%-ALE-60%	93.50%	52.80M(57.9%)	0.37M(56.5%)
ResNet-110	93.50%	252.89M(0.0%)	1.72M(0.0%)
ALE-80%	94.92%	120.29M(52.4%)	0.89M(48.3%)
20%-ALE-80%	94.55%	131.50M(48.0%)	0.96M(44.2%)
MobileNetV2	92.25%	91.15M(0.0%)	2.26M(0.0%)
ALE-90%	93.54%	40.86M(55.2%)	0.94M(58.4%)
20%- ALE-90%	94.11%	46.20M(49.3%)	1.08M(52.2%)

表12 嫁接网络个数的影响

模型	嫁接个数	准确率
ResNet-110		93.50%
ALE-80%	1	93.39%
	2	93.86%
	3	94.55%
	4	94.21%
	5	94.71%
	6	94.92%

表13 是否扩充信息熵实验对比

模型	准确率	计算量(压缩率)	参数量 (压缩率)
ResNet-56	93.26%	125.49M(0.0%)	0.85M(0.0%)
ALE-100%(扩)	94.26%	80.10M(36.2%)	0.54M(32.9%)
ALE-100%	94.11%	81.47M(35.1%)	0.59M(30.6%)
ALE-60%(扩)	93.64%	49.53M(60.5%)	0.36M(57.6%)
ALE-60%	93.37%	49.38M(60.7%)	0.36M(57.6%)
ResNet-110	93.50%	252.89M(0.0%)	1.72M(0.0%)
ALE-80%(扩)	94.92%	120.29M(52.4%)	0.89M(48.3%)
ALE-80%	94.81%	120.88M(52.2%)	0.91M(47.1%)
MobileNetV2	92.25%	91.15M(0.0%)	2.26M(0.0%)
ALE-90%(扩)	93.54%	40.86M(55.2%)	0.94M(58.4%)
ALE-90%	93.42%	38.90M(57.3%)	0.89M(60.6%)

参考文献 36

1	ZHU F, ZHU L, YANG Y. Sim-real joint reinforcement transfer for 3d indoor navigation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. California: IEEE, 2019: 11388-11397.
2	权宇, 李志欣, 张灿龙, 等. 融合深度扩张网络和轻量化网络的目标检测模型[J]. 电子学报, 2020, 48(2): 390-397.
	QUAN Y, LI Z X, ZHANG C L, et al. Fusing deep dilated convolutions network and light-weight network for object detection[J]. Acta Electronica Sinica, 2020, 48(2): 390-397. (in Chinese)
3	周涛, 霍兵强, 陆惠玲, 等. 残差神经网络及其在医学图像处理中的应用研究[J]. 电子学报, 2020, 48(7): 1436-1447.
	ZHOU T, HUO B Q, LU H L, et al. Research on residual neural network and its application on medical image processing[J]. Acta Electronica Sinica, 2020, 48(7): 1436-1447. (in Chinese)
4	DING X, DING G, GUO Y, et al. Centripetal sgd for pruning very deep convolutional networks with complicated structure[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. California: IEEE, 2019: 4943-4953.
5	饶川, 陈靓影, 徐如意, 等. 一种基于动态量化编码的深度神经网络压缩方法[J]. 自动化学报, 2019, 45(10):1960-1968.
	RAO C, CHEN J Y, XU R Y, et al. A dynamic quantization coding based deep neural network compression method[J]. Acta Automatica Sinica, 2019, 45(10): 1960-1968. (in Chinese)
6	ADRIANA R, NICOLAS B, SAMIRA E K, et al. Fitnets: Hints for thin deep nets[C]//International Conference on Learning Representations. California: OpenReview.net, 2015: 1-13.
7	ZHANG X, ZOU J, HE K, et al. Accelerating very deep convolutional networks for classification and detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 38(10): 1943-1955.
8	WANG Y, XU C, XU C, et al. Beyond filters: Compact feature map for portable deep model[C]//International Conference on Machine Learning. Sydney: ACM, 2017: 3703-3711.
9	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770-778.
10	SZEGEDY C, LIU W, JIA Y, et al. Going deeper with convolutions[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Boston: IEEE, 2015: 1-9.
11	CARREIRA-PERPINAN M A, IDELBAYEV Y. Learning-compression algorithms for neural net pruning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 8532-8541.
12	HAN S, POOL J, TRAN J, et al. Learning both weights and connections for efficient neural network[C]//Advances in Neural Information Processing Systems. Quebec: MIT Press, 2015: 1135-1143.
13	WEN W, WU C, WANG Y, et al. Learning structured sparsity in deep neural networks[C]//Advances in Neural Information Processing Systems. Barcelona: MIT Press, 2016: 2074-2082.
14	LUO J H, WU J, LIN W. Thinet: A filter level pruning method for deep neural network compression[C]//Proceedings of the IEEE International Conference on Computer Vision. Venice: IEEE, 2017: 5058-5066.
15	WANG D, ZHOU L, ZHANG X, et al. Exploring Linear Relationship in Feature Map Subspace for Convnets Compression[EB/OL]. (2018-03-15)[2020-12-01]. .
16	LI H, KADAV A, DURDANOVIC I, et al. Pruning filters for efficient convnets[C]//International Conference on Learning Representations. Toulon: OpenReview.net, 2017: 1-13.
17	YE J, LU X, LIN Z, et al. Rethinking the smaller-norm-less-informative assumption in channel pruning of convolution layers[C]//International Conference on Learning Representations. Vancouver: OpenReview.net, 2018: 1-11.
18	ZHUO H, QIAN X, FU Y, et al. Scsp: Spectral Clustering Filter Pruning with Soft Self-Adaption Manners[EB/OL]. (2018-07-14)[2020-12-01]. .
19	HE Y, LIU P, WANG Z, et al. Filter pruning via geometric median for deep convolutional neural networks acceleration[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. California: IEEE, 2019: 4340-4349.
20	LIU Z, SUN M, ZHOU T, et al. Rethinking the value of network pruning[C]//International Conference on Learning Representations. New Orleans: OpenReview.net, 2019: 1-21.
21	LIN S, JI R, YAN C, et al. Towards optimal structured cnn pruning via generative adversarial learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. California: IEEE, 2019: 2790-2799.
22	LIN M, JI R, ZHANG Y, et al. Channel pruning via automatic structure search[C]//International Joint Conference on Artificial Intelligence. Yokohama: Morgan Kaufmann, 2020: 1-7.
23	LIU Z, ZHANG X, SHEN Z, et al. Joint Multi-Dimension Pruning[EB/OL]. (2020-05-18)[2020-12-01]. .
24	MENG F, CHENG H, LI K, et al. Filter grafting for deep neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 6599-6607.
25	LUO J H, WU J. An Entropy-Based Pruning Method for CNN Compression[EB/OL]. (2017-07-19)[2020-12-01]. .
26	LI Y, LIN S, ZHANG B, et al. Exploiting kernel sparsity and entropy for interpretable CNN compression[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. California: IEEE, 2019: 2800-2809.
27	WANG L, XIONG Y, WANG Z, et al. Temporal segment networks: Towards good practices for deep action recognition[C]//Proceedings of the European Conference on Computer Vision, Amsterdam. Holland: Springer, 2016: 20-36.
28	KRIZHEVSKY A, HINTON G. Learning multiple layers of features from tiny images[J]. Technical Report, 2009, 1(1): 1-60.
29	NETZER Y, WANG T, COATES A, et al. Reading digits in natural images with unsupervised feature learning[J]. NIPS Workshop on Deep Learning and Unsupervised Feature Learning, 2011, 1(1): 1-9.
30	ZHENG L, SHEN L, TIAN L, et al. Scalable person re-identification: A benchmark[C]//Proceedings of the IEEE International Conference on Computer Vision. Santiago: IEEE, 2015: 1116-1124.
31	RISTANI E, SOLERA F, ZOU R, et al. Performance measures and a data set for multi-target, multi-camera tracking[C]//Proceedings of the European Conference on Computer Vision. Amsterdam: Springer, 2016: 17-35.
32	HUANG Z, WANG N. Data-driven sparse structure selection for deep neural networks[C]//Proceedings of the European Conference on Computer Vision. Munich: Springer, 2018: 304-320.
33	LIN M, JI R, WANG Y, et al. HRank: Filter pruning using high-rank feature map[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 1529-1538.
34	CHIN T W, DING R, ZHANG C, et al. Towards efficient model compression via learned global ranking[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 1518-1528.
35	YU R, LI A, CHEN C F, et al. Nisp: Pruning networks using neuron importance score propagation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 9194-9203.
36	LI Y, GU S, MAYER C, et al. Group sparsity: The hinge between filter pruning and decomposition for network compression[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 8018-8027.

[1]	黄赟, 张帆, 郭威, 陈立, 羊光. 一种基于数据标准差的卷积神经网络量化方法[J]. 电子学报, 2023, 51(3): 639-647.
[2]	吕杭, 蒋明峰, 李杨, 张鞠成, 王志康. 基于混合时频域特征的卷积神经网络心律失常分类方法的研究[J]. 电子学报, 2023, 51(3): 701-711.
[3]	许新征, 李杉. 基于特征膨胀卷积模块的轻量化技术研究[J]. 电子学报, 2023, 51(2): 355-364.
[4]	丁琪, 田萱, 孙国栋. 基于注意力增强的热点感知新闻推荐模型[J]. 电子学报, 2023, 51(1): 93-104.
[5]	张永梅, 孙捷. 基于动静态特征双输入神经网络的咳嗽声诊断COVID-19算法[J]. 电子学报, 2023, 51(1): 202-212.
[6]	袁海英, 成君鹏, 曾智勇, 武延瑞. Mobile_BLNet：基于Big-Little Net的轻量级卷积神经网络优化设计[J]. 电子学报, 2023, 51(1): 180-191.
[7]	吴靖, 叶晓晶, 黄峰, 陈丽琼, 王志锋, 刘文犀. 基于深度学习的单帧图像超分辨率重建综述[J]. 电子学报, 2022, 50(9): 2265-2294.
[8]	毛国君, 王者浩, 黄山, 王翔. 基于剪边策略的图残差卷积深层网络模型[J]. 电子学报, 2022, 50(9): 2205-2214.
[9]	袁海英, 曾智勇, 成君鹏. 面向灵活并行度的稀疏卷积神经网络加速器[J]. 电子学报, 2022, 50(8): 1811-1818.
[10]	张志文, 刘天歌, 聂鹏举. 基于实景数据增强和双路径融合网络的实时街景语义分割算法[J]. 电子学报, 2022, 50(7): 1609-1620.
[11]	王相海, 赵晓阳, 王鑫莹, 赵克云, 宋传鸣. 非抽取小波边缘学习深度残差网络的单幅图像超分辨率重建[J]. 电子学报, 2022, 50(7): 1753-1765.
[12]	张波, 陆云杰, 秦东明, 邹国建. 一种卷积自编码深度学习的空气污染多站点联合预测模型[J]. 电子学报, 2022, 50(6): 1410-1427.
[13]	丁毅, 沈薇, 李海生, 钟琼慧, 田明宇, 李洁. 面向CNN的区块链可信隐私服务计算模型[J]. 电子学报, 2022, 50(6): 1399-1409.
[14]	张云, 化青龙, 姜义成, 徐丹. 基于混合型复数域卷积神经网络的三维转动舰船目标识别[J]. 电子学报, 2022, 50(5): 1042-1049.
[15]	廖勇, 李玉杰. 一种轻量化低复杂度的FDD大规模MIMO系统CSI反馈方法[J]. 电子学报, 2022, 50(5): 1211-1217.