«上一篇
 文章快速检索 高级检索

 智能系统学报  2019, Vol. 14 Issue (3): 566-574  DOI: 10.11992/tis.201804056 0

### 引用本文

YU Ying, WANG Lewei, WU Xinnian, et al. A multi-label classification algorithm based on an improved convolutional neural network[J]. CAAI Transactions on Intelligent Systems, 2019, 14(3): 566-574. DOI: 10.11992/tis.201804056.

### 文章历史

1. 华东交通大学 软件学院，江西 南昌 330013;
2. 中南大学 交通运输工程学院，湖南 长沙 410000;
3. 同济大学计算机科学与技术系，上海 201804

A multi-label classification algorithm based on an improved convolutional neural network
YU Ying 1, WANG Lewei 1, WU Xinnian 1, WU Guohua 2, ZHANG Yuanjian 3
1. College of Software Engineering, East China Jiaotong University, Nanchang 330013, China;
2. College of Transportation Engineering, Central South University, Changsha 410000, China;
3. Department of Computer Science and Technology, Tongji University, Shanghai 201804, China
Abstract: A good feature expression is the key to improve model performance. However, at present, artificially designed features are used for multi-label learning. Thus, the level of abstraction of the extracted features is low and lacks the discriminated information involved. To solve this problem, this paper proposes a multi-label classification model based on convolutional neural network (ML_DCCNN). This model uses the powerful feature extraction capabilities of CNNs to automatically learn the features from the data. To solve the problem of high forecasting precision versus long training time of CNNs, the ML_DCCNN uses the transfer learning method to reduce the training time of the model. In addition, the entire connection layer of the CNN is improved by a dual-channel neuron, which can reduce the number of parameters of the fully connected layer. The experiments show that compared with the traditional multi-label classification algorithm and existing multi-label classification model based on deep learning, the ML_DCCNN maintains high classification accuracy and can effectively improve the classification efficiency, presenting certain theoretical and practical value.
Key words: multi-label learning    convolutional neural network    transfer learning    fully-connected layer    feature expression    multi-label classification    deep learning    loss function

1 相关工作 1.1 多标记学习

1.2 卷积神经网络

1.3 迁移学习

2 基于改进CNN的多标记分类算法 2.1 算法框架

 Download: 图 2 基于改进CNN的多标记分类算法框架 Fig. 2 Multi-label classification algorithm framework based on improved convolution neural network
2.2 双通道神经元

2.2.1 基本结构

 $\frac{1}{2} \leqslant \frac{{\left( {m + 1} \right)\left( {d + e} \right) + 2d}}{{\left( {m + 1} \right)n}} \leqslant 1$ (1)

2.2.2 核心思想

2.3 损失函数

$D = \{ ({{{x}}_i},{{{y}}_i})|i = 1,2, \cdots, n\}$ 代表具有 $n$ 个样本的训练集，其中 ${{{x}}_i} = [{x_{i1}}\;{x_{i2}} \cdots {x_{id}}]$ 是第 $i$ 个样本的 $d$ 维特征向量， ${{{y}}_i} = [{y_{i1}}\;{y_{i2}} \cdots {y_{iq}}]$ 是第 $i$ 个样本的标记向量，其维度 $q$ 与数据集标记总数相等， ${y_i}_j = 1$ 表示xi含有标签 ${l_j}$ ${y_{ij}} = 0$ 则表示不含有。

SoftMax分类器不仅可以用于处理单标记分类问题，也可以用于处理多标记分类问题。本文将最后一层全连接层的输出送入SoftMax分类器中，得出图片含有各标记的概率，例如图片xi含有标记 ${l_j}$ 的概率：

 ${p_{ij}}{\rm{ = }}\frac{{\exp ({f_j}({{{x}}_i}))}}{{\sum\limits_{k = 1}^c {\exp ({f_k}({{{x}}_i}))} }}$ (2)

 $J = - \sum\limits_{i = 1}^n {\sum\limits_{j = 1}^q {\overline {{p_{ij}}} \log ({p_{ij}})} }$ (3)

 $\overline {{p_{ij}}} {\rm{ = }}\left\{ {\begin{array}{*{20}{c}} {\displaystyle\frac{1}{{{c_ + }}}},&{ {{y_{ij}} = 1} } \\ 0,&{ {{y_{ij}} = 0} } \end{array}} \right.$ (4)

 $J = - \sum\limits_{i = 1}^n {\sum\limits_{j = 1}^{{c_ + }} {\frac{1}{{{c_ + }}}\log ({p_{ij}})} }$ (5)

3 实验与分析

 Download: 图 6 双通道神经元比例 $\lambda$ 对平均准确率mAP和参数缩 减比例 $\beta$ 的影响 Fig. 6 Effect of dual-channel neuron ratio $\lambda$ on the mAP and ratio $\beta$ of parameter reduction

4 结束语

 [1] TROHIDIS K, TSOUMAKAS G, KALLIRIS G, et al. Multilabel classification of music into emotions[C]//Proceedings of 2008 International Conference on Music Information Retrieval (ISMIR 2008). Philadelphia, USA, 2008: 325–330. (0) [2] WU Baoyuan, LYU S, HU Baoguang, et al. Multi-label learning with missing labels for image annotation and facial action unit recognition[J]. Pattern recognition, 2015, 48(7): 2279-2289. DOI:10.1016/j.patcog.2015.01.022 (0) [3] JIANG J Q, MCQUAY L J. Predicting protein function by multi-label correlated semi-supervised learning[J]. IEEE/ACM transactions on computational biology and bioinformatics, 2012, 9(4): 1059-1069. DOI:10.1109/TCBB.2011.156 (0) [4] OZONAT K, YOUNG D. Towards a universal marketplace over the web: statistical multi-label classification of service provider forms with simulated annealing[C]//Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Paris, France, 2009: 1295–1304. (0) [5] GONG Y, JIA Y, LEUNG T, et al. Deep convolutional ranking for multilabel image annotation[C]. 2nd International Conference on Learning Representations, ICLR 2014. Banff, Canada, 2014: 1312–1320. (0) [6] ZHANG Minling, ZHOU Zhihua. A review on multi-label learning algorithms[J]. IEEE transactions on knowledge and data engineering, 2014, 26(8): 1819-1837. DOI:10.1109/TKDE.2013.39 (0) [7] LUACES O, DÍEZ J, BARRANQUERO J, et al. Binary relevance efficacy for multilabel classification[J]. Progress in artificial intelligence, 2012, 1(4): 303-313. DOI:10.1007/s13748-012-0030-x (0) [8] READ J, PFAHRINGER B, HOLMES G. Multi-label classification using ensembles of pruned sets[C]//ICDM'08. Eighth IEEE International Conference on Data Mining. Pisa, Italy, 2008: 995–1000. (0) [9] WAN Shupeng, XU Jianhua. A multi-label classification algorithm based on triple class support vector machine[C]//Proceedings of 2007 International Conference on Wavelet Analysis and Pattern Recognition. Beijing, China, 2008: 1447–1452. (0) [10] 张敏灵. 一种新型多标记懒惰学习算法[J]. 计算机研究与发展, 2012, 49(11): 2271-2282. ZHANG Minling. An improved multi-label lazy learning approach[J]. Journal of computer research and development, 2012, 49(11): 2271-2282. (0) [11] ZHANG Mimling. ML-RBF: RBF neural networks for multi-label learning[J]. Neural processing letters, 2009, 29(2): 61-74. DOI:10.1007/s11063-009-9095-3 (0) [12] WANG Jiang, YANG Yi, MAO Junhua, et al. CNN-RNN: a unified framework for multi-label image classification[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA, 2016: 2285–2294. (0) [13] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems. Lake Tahoe, USA, 2012: 1097–1105. (0) [14] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[C]//3rd International Conference on Learning Representations, ICLR 2015. San Diego, USA, 2015:1409–1422. (0) [15] DAI Wenyuan, XUE Guirong, YANG Qiang, et al. Co-clustering based classification for out-of-domain documents[C]//Proceedings of the 13th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Jose, USA, 2007: 210–219. (0) [16] ZHUO H H, YANG Qiang. Action-model acquisition for planning via transfer learning[J]. Artificial intelligence, 2014, 212: 80-103. DOI:10.1016/j.artint.2014.03.004 (0) [17] DENG Jia, DONG Wei, SOCHER R, et al. ImageNet: a large-scale hierarchical image database[C]//Proceedings of 2009 IEEE Conference on Computer Vision and Pattern Recognition. Miami, USA, 2009: 248–255. (0) [18] SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the Inception Architecture for Computer Vision[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA, 2016: 2818–2826. (0) [19] EVERINGHAM M, VAN GOOL L, WILLIAMS C K I, et al. The pascal visual object classes (VOC) challenge[J]. International journal of computer vision, 2010, 88(2): 303-338. DOI:10.1007/s11263-009-0275-4 (0) [20] SHELHAMER E, LONG J, DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 39(4): 640-651. DOI:10.1109/TPAMI.2016.2572683 (0) [21] ZHANG Chenlin, LUO Jianhao, WEI Xiushen, et al. In defense of fully connected layers in visual representation transfer[C]//Proceedings of the 18th Pacific-Rim Conference on Multimedia on Advances in Multimedia Information Processing. Harbin, China, 2017: 807–817. (0) [22] HARZALLAH H, JURIE F, SCHMID C. Combining efficient object localization and image classification[C]//Proceedings of 2009 IEEE International Conference on Computer Vision. Kyoto, Japan, 2009: 237–244. (0) [23] PERRONNIN F, SÁNCHEZ J, MENSINK T. Improving the fisher kernel for large-scale image classification[C]//Proceedings of the 11th European Conference on Computer Vision. Heraklion, Greece, 2010: 143–156. (0) [24] YANG Jingjing, LI Yuanning, TIAN Yonghong, et al. Group-sensitive multiple kernel learning for object categorization[C]//Proceedings of 2009 IEEE International Conference on Computer Vision. Kyoto, Japan, 2009: 436–443. (0) [25] OQUAB M, BOTTOU L, LAPTEV I, et al. Learning and transferring mid-level image representations using convolutional neural networks[C]//Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition. Columbus, USA, 2014: 1717–1724. (0) [26] WEI Y , XIA W , LIN M , et al. HCP: a flexible CNN framework for multi-label image classification[J]. IEEE transactions on pattern analysis and machine intelligence, 2016, 38(9): 1901–1907. (0)