《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (12): 3955-3964.DOI: 10.11772/j.issn.1001-9081.2022121873
• 前沿与综合应用 • 上一篇
收稿日期:
2022-12-21
修回日期:
2023-03-01
接受日期:
2023-03-08
发布日期:
2023-04-19
出版日期:
2023-12-10
通讯作者:
王正刚
作者简介:
王正刚(1984—),男,四川成都人,高级工程师,博士研究生,主要研究方向:计算机软件与理论、人工智能;Email:wangzhenggang@customs.gov.cn.com基金资助:
Zhenggang WANG1,2,3(), Zhong LIU1,2, Jin JIN4, Wei LIU3
Received:
2022-12-21
Revised:
2023-03-01
Accepted:
2023-03-08
Online:
2023-04-19
Published:
2023-12-10
Contact:
Zhenggang WANG
About author:
LIU Zhong, born in 1968, Ph. D., research fellow. His research interests include computer software and theory, machine certification.Supported by:
摘要:
针对现阶段我国海关风险布控方法存在效率、准确率较低、人力资源占用过多的问题和智能化分类算法小型化部署需求,提出一种基于改进蝶形反馈型神经网络(BFNet-V2)的海关风险布控方法。首先,运用编码填充(FC)算法实现海关表格数据到模拟图像的语义替换;其次,运用BFNet-V2训练模拟图像数据,由左右两条链路、不同卷积核和块、小块的设计组成规则的神经网络结构,并添加残差短路径干预改善过拟合和梯度消失;最后,提出历史动量自适应矩估计算法(H-Adam)优化梯度下降过程,取得更优的自适应学习率调整方式,并分类海关数据。选取Xception(eXtreme inception)、移动网络(MobileNet)、残差网络(ResNet)和蝶形反馈型神经网络(BF-Net)为基线网络结构进行对比。BFNet-V2的接受者工作特征曲线(ROC)和查准率-查全率曲线(PR)包含了基线网络结构的曲线,与4种基线网络结构相比,基于迁移学习(TL)的BFNet-V2分类准确率分别提高了4.30%、4.34%、4.10%和0.37%。在真实标签数据分类过程中,BFNet-V2的查获误判率分别降低了70.09%、57.98%、58.36%和10.70%。比较所提方法与包含浅层和深度学习方法在内的8种分类方法,在3个数据集上的准确率均提升1.33%以上,可见所提方法能够实现表格数据自动分类,提升海关风险布控的效率和准确度。
中图分类号:
王正刚, 刘忠, 金瑾, 刘伟. 基于改进蝶形反馈型神经网络的海关风险布控方法[J]. 计算机应用, 2023, 43(12): 3955-3964.
Zhenggang WANG, Zhong LIU, Jin JIN, Wei LIU. Customs risk control method based on improved butterfly feedback neural network[J]. Journal of Computer Applications, 2023, 43(12): 3955-3964.
标签 | 数据量 | 标签 | 数据量 |
---|---|---|---|
00 | 16 323 | 07 | 11 522 |
02 | 15 015 | 08 | 11 581 |
03 | 10 542 | 12 | 12 402 |
05 | 12 073 | 13 | 15 703 |
06 | 13 544 | 14 | 14 285 |
表1 数据集分布
Tab.1 Dataset distribution
标签 | 数据量 | 标签 | 数据量 |
---|---|---|---|
00 | 16 323 | 07 | 11 522 |
02 | 15 015 | 08 | 11 581 |
03 | 10 542 | 12 | 12 402 |
05 | 12 073 | 13 | 15 703 |
06 | 13 544 | 14 | 14 285 |
训练 方式 | 网络结构 | 指标平均值/% | 参数量 | |||
---|---|---|---|---|---|---|
ACC | AUC | AP | Kappa | |||
TS | Xception | 89.33 | 92.37 | 89.33 | 87.69 | 23 177 538 |
MobileNet-V2 | 6 601 954 | |||||
ResNet50 | 88.67 | 91.58 | 89.19 | 87.91 | 44 984 002 | |
BF-Net | 92.81 | 97.46 | 94.84 | 91.23 | 7 790 124 | |
BFNet-V2 | 93.27 | 98.24 | 94.98 | 91.89 | 7 294 320 | |
TL | Xception | 90.10 | 93.01 | 91.22 | 88.51 | |
MobileNet-V2 | 92.06 | 88.13 | ||||
ResNet50 | 90.27 | 90.63 | ||||
BF-Net | 93.62 | 97.94 | 95.03 | 92.73 | ||
BFNet-V2 | 93.97 | 98.25 | 95.84 | 93.24 |
表2 验证数据集指标平均值
Tab.2 Mean index values on validation dataset
训练 方式 | 网络结构 | 指标平均值/% | 参数量 | |||
---|---|---|---|---|---|---|
ACC | AUC | AP | Kappa | |||
TS | Xception | 89.33 | 92.37 | 89.33 | 87.69 | 23 177 538 |
MobileNet-V2 | 6 601 954 | |||||
ResNet50 | 88.67 | 91.58 | 89.19 | 87.91 | 44 984 002 | |
BF-Net | 92.81 | 97.46 | 94.84 | 91.23 | 7 790 124 | |
BFNet-V2 | 93.27 | 98.24 | 94.98 | 91.89 | 7 294 320 | |
TL | Xception | 90.10 | 93.01 | 91.22 | 88.51 | |
MobileNet-V2 | 92.06 | 88.13 | ||||
ResNet50 | 90.27 | 90.63 | ||||
BF-Net | 93.62 | 97.94 | 95.03 | 92.73 | ||
BFNet-V2 | 93.97 | 98.25 | 95.84 | 93.24 |
网络结构 | 训练时间 |
---|---|
Xception | 31.289 9 |
MobileNet-V2 | 16.821 4 |
ResNet50 | 24.727 7 |
BF-Net | 22.232 9 |
BFNet-V2 | 19.453 2 |
表3 基于迁移学习的网络训练时间对比 (s)
Tab.3 Comparison of network training time based on transfer learning
网络结构 | 训练时间 |
---|---|
Xception | 31.289 9 |
MobileNet-V2 | 16.821 4 |
ResNet50 | 24.727 7 |
BF-Net | 22.232 9 |
BFNet-V2 | 19.453 2 |
误判 种类 | 训练方式 | 网络结构 | ||||
---|---|---|---|---|---|---|
Xception | MobileNet-V2 | ResNet50 | BF-Net | BFNet-V2 | ||
放行 误判 | TS | 10.030 | 9.683 | 7.983 | 2.212 | 2.034 |
TL | 5.992 | 7.865 | 6.742 | 1.997 | 1.871 | |
查获 误判 | TS | 4.328 | 3.415 | 3.511 | 1.425 | 1.235 |
TL | 3.795 | 2.701 | 2.726 | 1.271 | 1.135 |
表4 5种网络结构的误判率对比 ( %)
Tab.4 Comparison of misjudgment rate among five network structures
误判 种类 | 训练方式 | 网络结构 | ||||
---|---|---|---|---|---|---|
Xception | MobileNet-V2 | ResNet50 | BF-Net | BFNet-V2 | ||
放行 误判 | TS | 10.030 | 9.683 | 7.983 | 2.212 | 2.034 |
TL | 5.992 | 7.865 | 6.742 | 1.997 | 1.871 | |
查获 误判 | TS | 4.328 | 3.415 | 3.511 | 1.425 | 1.235 |
TL | 3.795 | 2.701 | 2.726 | 1.271 | 1.135 |
方法 | 数据集分类准确率 | ||
---|---|---|---|
Adult | LETOR4.0 | Cardiovascular | |
RF[ | 79.7 | 76.7 | 78.6 |
SVM[ | 80.9 | 79.9 | 80.2 |
XGBoost[ | 82.2 | 81.2 | 82.4 |
1D-CNN[ | 85.5 | 82.5 | 85.1 |
DANET[ | 86.5 | 85.9 | 84.5 |
TAC[ | 87.8 | 85.8 | 86.7 |
SuperTML[ | 87.9 | 86.9 | 85.3 |
BF-Net+AdaGrad[ | 90.2 | 89.2 | 88.7 |
BFNet-V2+H-Adam | 91.4 | 90.8 | 89.9 |
表5 不同方法的准确率对比 ( %)
Tab. 5 Accuracy comparison of different methods
方法 | 数据集分类准确率 | ||
---|---|---|---|
Adult | LETOR4.0 | Cardiovascular | |
RF[ | 79.7 | 76.7 | 78.6 |
SVM[ | 80.9 | 79.9 | 80.2 |
XGBoost[ | 82.2 | 81.2 | 82.4 |
1D-CNN[ | 85.5 | 82.5 | 85.1 |
DANET[ | 86.5 | 85.9 | 84.5 |
TAC[ | 87.8 | 85.8 | 86.7 |
SuperTML[ | 87.9 | 86.9 | 85.3 |
BF-Net+AdaGrad[ | 90.2 | 89.2 | 88.7 |
BFNet-V2+H-Adam | 91.4 | 90.8 | 89.9 |
1 | ALITA D, PUTRA A D, DARWIS D. Analysis of classic assumption test and multiple linear regression coefficient test for employee structural office recommendation[J]. IJCCS (Indonesian Journal of Computing and Cybernetics Systems), 2021, 15(3): 295-306. 10.22146/ijccs.65586 |
2 | CHARBUTY B, ABDULAZEEZ A. Classification based on decision tree algorithm for machine learning[J]. Journal of Applied Science and Technology Trends, 2021, 2(1): 20-28. 10.38094/jastt20165 |
3 | HUSSEIN A S, KHAIRY R S, NAJEEB S M M, et al. Credit card fraud detection using fuzzy rough nearest neighbor and sequential minimal optimization with logistic regression[J]. International Journal of Interactive Mobile Technologies, 2021, 15(5): 24-42. 10.3991/ijim.v15i05.17173 |
4 | ANTONIADIS A, LAMBERT-LACROIX S, J-M POGGI. Random forests for global sensitivity analysis: a selective review[J]. Reliability Engineering & System Safety, 2021, 206: 107312. 10.1016/j.ress.2020.107312 |
5 | LIAW A, WIENER M. Classification and regression by random forest[J]. R News, 2002,2(3): 18-22. |
6 | 申明尧, 韩萌, 杜诗语,等. 融合XGBoost和Multi-GRU的数据中心服务器能耗优化算法[J]. 计算机应用, 2022, 42(1): 198-208. 10.11772/j.issn.1001-9081.2021071291 |
SHEN M Y, HAN M, DU S Y, et al. Data center server energy consumption optimization algorithm combining XGBoost and Multi-GRU [J]. Journal of Computer Applications, 2022, 42(1): 198-208. 10.11772/j.issn.1001-9081.2021071291 | |
7 | CHEN J, LIAO K, WANY, et al. DANETs: deep abstract networks for tabular data classification and regression[C]// Proceedings of the 36th AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2022: 3930-3938. 10.1609/aaai.v36i4.20309 |
8 | BUTUROVIĆ L, MILJKOVIC D. A novel method for classification of tabular data using convolutional neural networks [EB/OL]. (2020-03-08) [2023-01-12]. . 10.1101/2020.05.02.074203 |
9 | SUN B, YANG L, ZHANG W, et al. SuperTML: two-dimensional word embedding for the precognition on structured tabular data [C]// Proceedings of the 32th IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Piscataway: IEEE, 2019: 2973-2981. 10.1109/cvprw.2019.00360 |
10 | CHOLLET F. Xception: deep learning with depth wise separable convolutions[C]// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017: 1800-1807. 10.1109/cvpr.2017.195 |
11 | SANDLER M, HOWARD A, ZHU M, et al. MobileNetV2: inverted residuals and linear bottlenecks [C]// Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 4510-4520. 10.1109/cvpr.2018.00474 |
12 | WANG G, YU H, SUI Y. Research on maize disease recognition method based on improved ResNet50[J]. Mobile Information Systems, 2021, 2021: 9110866.1-9110866.6. 10.1155/2021/9110866 |
13 | 王正刚, 刘伟, 金瑾.一种海关数据风控类型识别方法, 海关智能化风险布控方法, 装置, 计算机设备及存储介质: CN202110232188.2[P]. 2022-09-16. |
WANG Z G, LIU W, JIN J. A customs data risk control type identification method, customs intelligent risk control method, device, computer equipment and storage media: CN202110232188.2 [P]. 2022-09-16. | |
14 | WEI Y, XIAO H, SHI H, et al. Revisiting dilated convolution: a simple approach for weakly-and semi-supervised semantic segmentation [C]// Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 7268-7277. 10.1109/cvpr.2018.00759 |
15 | DAUBECHIES I, DeVORE R, FOUCART S, et al. Nonlinear approximation and (deep) ReLU networks[J]. Constructive Approximation: An International Journal for Approximations and Expansions, 2022, 55(1): 127-172. 10.1007/s00365-021-09548-z |
16 | JAIS I K M, ISMAILI A R, NISA S Q. Adam optimization algorithm for wide and deep neural network[J]. Knowledge Engineering and Data Science, 2019, 2(1): 41-46. 10.17977/um018v2i12019p41-46 |
17 | HAWKINS D M, YOUNG S S, Ⅲ RUSINKO A. Analysis of a large structure-activity data set using recursive partitioning[J]. Quantitative Structure-Activity Relationships, 1997, 16(4):296-302. 10.1002/qsar.19970160404 |
18 | WARD R, WU X, BOTTOU L. AdaGrad stepsizes: sharp convergence over nonconvex landscapes, from any initialization[J]. Journal of Machine Learning Research, 2020, 21: 1-30. |
19 | WEN Z, YANG G, CAI Q. An improved calibration method for the IMU biases utilizing KF-based AdaGrad algorithm[J]. Sensors, 2021, 21(15): 5055. 10.3390/s21155055 |
20 | LI L, XU W, YU H. Character-level neural network model based on Nadam optimization and its application in clinical concept extraction [J]. Neurocomputing, 2020, 414: 182-190. 10.1016/j.neucom.2020.07.027 |
21 | ZHU Z, HOU Z. Research and application of rectified-nadam optimization algorithm in data classification [J]. American Journal of Computer Science and Technology, 2021, 4(4): 106-110. 10.11648/j.ajcst.20210404.13 |
22 | GU J, WANG Z, KUEN J, et al. Recent advances in convolutional neural networks [J]. Pattern Recognition, 2018,77: 354-377. 10.1016/j.patcog.2017.10.013 |
23 | YU S, CHENG Y, SU S, et al. Stratified pooling based deep convolutional neural networks for human action recognition[J]. Multimedia Tools and Applications, 2017, 76: 13367-13382. 10.1007/s11042-016-3768-5 |
24 | KIM Y, PANDA P. Revisiting batch normalization for training low-latency deep spiking neural networks from scratch [J]. Frontiers in Neuroscience, 2021,15: 101-113. 10.3389/fnins.2021.773954 |
25 | KARRAS T, AITTALA M, HELLSTEN J, et al. Training generative adversarial networks with limited data [J]. Advances in Neural Information Processing Systems, 2020, 33: 12104-12114. |
26 | SHALLU, MEHRA R. Breast cancer histology images classification: training from scratch or transfer learning [J]. ICT Express, 2018, 4(4): 247-254. 10.1016/j.icte.2018.10.007 |
27 | DOKUZ Y, TUFEKCI Z. Mini-batch sample selection strategies for deep learning based speech recognition[J]. Applied Acoustics, 2021, 171: 107573. 10.1016/j.apacoust.2020.107573 |
28 | ZHANG Z, SABUNCU M. Generalized cross entropy loss for training deep neural networks with noisy labels [EB/OL]. (2018-07-15) [2022-12-25]. . 10.48550/arXiv.1805.07836 |
29 | DENG J, DONG W, SOCHER R, et al. ImageNet: a largescale hierarchical image database[C]// Proceedings of the 22th IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2009: 248-255. 10.1109/cvpr.2009.5206848 |
30 | PANG B, NIJKAMP E, WU Y N. Deep learning with TensorFlow: a review [J]. Journal of Educational and Behavioral Statistics, 2020, 45(2): 227-248. 10.3102/1076998619872761 |
31 | YANG J, SUN L, XING W, et al. Hyperspectral prediction of sugarbeet seed germination based on Gauss kernel SVM[J]. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2021, 253: 119585. 10.1016/j.saa.2021.119585 |
32 | OZCANLI A K, BAYSAL M. Islanding detection in microgrid using deep learning based on 1D-CNN and CNN-LSTM networks[J]. Sustainable Energy, Grids and Networks, 2022, 32: 100839. 10.1016/j.segan.2022.100839 |
33 | RONNY K, BARRY B. Adult data set [DB/OL]. (2003-06-15)[2022-12-14] . . |
34 | PENG J, MACDONALD C, OUNIS I. Learning to select a ranking function [C]// Proceedings of the 32th European Conference on IR Research. Berlin: Springer, 2010: 114-126. 10.1007/978-3-642-12275-0_13 |
35 | ULIANOVA S. Cardiovascular disease dataset[DB/OL]. (2005-03-08)[2022-12-14]. . 10.1037/e409612008-001 |
[1] | 秦璟, 秦志光, 李发礼, 彭悦恒. 基于概率稀疏自注意力神经网络的重性抑郁疾患诊断[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2970-2974. |
[2] | 李云, 王富铕, 井佩光, 王粟, 肖澳. 基于不确定度感知的帧关联短视频事件检测方法[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2903-2910. |
[3] | 陈虹, 齐兵, 金海波, 武聪, 张立昂. 融合1D-CNN与BiGRU的类不平衡流量异常检测[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2493-2499. |
[4] | 赵宇博, 张丽萍, 闫盛, 侯敏, 高茂. 基于改进分段卷积神经网络和知识蒸馏的学科知识实体间关系抽取[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2421-2429. |
[5] | 张春雪, 仇丽青, 孙承爱, 荆彩霞. 基于两阶段动态兴趣识别的购买行为预测模型[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2365-2371. |
[6] | 王东炜, 刘柏辰, 韩志, 王艳美, 唐延东. 基于低秩分解和向量量化的深度网络压缩方法[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 1987-1994. |
[7] | 高阳峄, 雷涛, 杜晓刚, 李岁永, 王营博, 闵重丹. 基于像素距离图和四维动态卷积网络的密集人群计数与定位方法[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2233-2242. |
[8] | 沈君凤, 周星辰, 汤灿. 基于改进的提示学习方法的双通道情感分析模型[J]. 《计算机应用》唯一官方网站, 2024, 44(6): 1796-1806. |
[9] | 黄梦源, 常侃, 凌铭阳, 韦新杰, 覃团发. 基于层间引导的低光照图像渐进增强算法[J]. 《计算机应用》唯一官方网站, 2024, 44(6): 1911-1919. |
[10] | 李健京, 李贯峰, 秦飞舟, 李卫军. 基于不确定知识图谱嵌入的多关系近似推理模型[J]. 《计算机应用》唯一官方网站, 2024, 44(6): 1751-1759. |
[11] | 姚迅, 秦忠正, 杨捷. 生成式标签对抗的文本分类模型[J]. 《计算机应用》唯一官方网站, 2024, 44(6): 1781-1785. |
[12] | 高文烁, 陈晓云. 基于节点结构的点云分类网络[J]. 《计算机应用》唯一官方网站, 2024, 44(5): 1471-1478. |
[13] | 席治远, 唐超, 童安炀, 王文剑. 基于双路时空网络的驾驶员行为识别[J]. 《计算机应用》唯一官方网站, 2024, 44(5): 1511-1519. |
[14] | 孙敏, 成倩, 丁希宁. 基于CBAM-CGRU-SVM的Android恶意软件检测方法[J]. 《计算机应用》唯一官方网站, 2024, 44(5): 1539-1545. |
[15] | 陈天华, 朱家煊, 印杰. 基于注意力机制的鸟类识别算法[J]. 《计算机应用》唯一官方网站, 2024, 44(4): 1114-1120. |
阅读次数 | ||||||
全文 |
|
|||||
摘要 |
|
|||||