近年来，互联网的飞速发展和数码相机、电脑及智能手机等工具的普及使用，导致图像信息爆炸式增长，如何对这些图像进行分类也便成为了一个亟待解决的问题。像自然场景这样较复杂的图像，传统的监督学习框架已经不能满足所需要求，学者们又相继提出了多示例学习框架、多标记学习框架和多示例多标记（MIML）学习框架。

本文主要是将多示例多标记学习框架应用到自然场景图像的分类中，展开了对多示例多标记学习算法的研究学习，并针对基于RBF神经网络的MIML算法进行了改进，本文的主要工作如下：

（1）MIML学习算法处理的是包含多个示例的包，包生成方法的好坏直接影响到最终的分类结果。本文研究并实现了基于固定区域和基于图像分割的两类包生成方法，通过实验证明，在自然场景图像分类中，SBN（Single Blob with Neighbors）包生成方法效果较好。

（2）针对自然场景图像分类，研究并实现了MIMLBOOST、MIMLSVM、M³MIML和MIML-KNN等算法：MIMLBOOST和MIMLSVM是基于退化策略的，它们假设示例间或标签间是相互独立的，损失了很多的有用信息；而M³MIML和MIML-KNN则考虑了示例和标签的相关性信息。通过实验表明，考虑了相关性信息的M³MIML和MIML-KNN算法在分类效果上要优于未考虑相关性信息的MIMLBOOST和MIMLSVM算法。

（3）基于神经网络对MIML展开了进一步研究，主要分析了基于RBF神经网络的MIML算法，并针对该算法进行改进——为平均Hausdorff距离引入一个自动调节系数。通过实验表明：①针对自然场景图像分类的MIML算法中，基于RBF网络的算法要优于基于BP网络，也优于其他传统算法；②在最大Hausdorff、最小Hausdorff、平均Hausdorff和三种Hausdorff距离均值这四种度量方式中，采用平均Hausdorff距离的分类效果最好；③针对平均Hausdorff距离削弱了两包之间最近示例距离所起作用的问题，改进之后的平均Hausdorff距离可以进一步提高分类的精度。

[1] Hirata K, Kato T. Query by visual example[C].Advances in Database Technology—EDBT’92. 1992: 56–71.

[2] Dietterich T G, Lathrop R H, Lozano-perez T. Solving the multiple instance problem with axis-parallel rectangles[J]. Artificial intelligence, Elsevier, 1997, 89(1): 31–71.

[3] Zhou Z-H, Zhang M-L. Multi-instance multi-label learning with application to scene classification[C]. Advances in neural information processing systems. 2006: 1609–1616.

[4] 段震. 基于构造性学习的覆盖算法的发展及应用[D]. 安徽大学, 2010.

[5] 李大湘, 赵小强, 李娜. 图像语义分析的多示例学习算法综述[J]. 控制与决策, 2013, 28(4): 481–488.

[6] Maron O, Ratan A L. Multiple-Instance Learning for Natural Scene Classification[C]. ICML. 1998, 98: 341–349.

[7] Zhang Q, Goldman S A. EM-DD: An improved multiple-instance learning technique[C]. Advances in neural information processing systems. 2001: 1073–1080.

[8] Mason L, Baxter J, Bartlett P, et al. Boosting algorithms as gradient descent in function space[C].1999.

[9] Andrews S, Tsochantaridis I, HOFMANN T. Support vector machines for multiple-instance learning[C]. Advances in neural information processing systems. 2002: 561–568.

[10] Gehler P V, Chapelle O. Deterministic annealing for multiple-instance learning[C]. International conference on artificial intelligence and statistics. 2007: 123–130.

[11] Chen Y, Wang J Z. Image categorization by learning and reasoning with regions[J]. The Journal of Machine Learning Research, JMLR. org, 2004, 5: 913–939.

[12] Chen Y, Bi J, Wang J Z. MILES: Multiple-instance learning via embedded instance selection[J]. Pattern Analysis and Machine Intelligence, IEEE Transactions on, IEEE, 2006, 28(12): 1931–1947.

[13] Zhang M-L, Zhang K. Multi-label learning by exploiting label dependency[C].Proceedings of the 16th ACM SIGKDD international conference on Knowledge discovery and data mining. 2010: 999–1008.

[14] Huang S-J, Zhou Z-H, Zhou Z H. Multi-Label Learning by Exploiting Label Correlations Locally.[C].AAAI. 2012.

[15] 李志欣, 卓亚琦, 张灿龙, et al. 多标记学习研究综述[J]. 计算机应用研究, 2014, 31(6): 1601–1605.

[16] Chen W, Yan J, Zhang B, et al. Document transformation for multi-label feature selection in text categorization[C]. Data Mining, 2007. ICDM 2007. Seventh IEEE International Conference on. 2007: 451–456.

[17] Clare A, King R D. Knowledge discovery in multi-label phenotype data[G]. Principles of data mining and knowledge discovery. Springer, 2001: 42–53.

[18] Zhang M-L, Zhou Z-H. Multilabel neural networks with applications to functional genomics and text categorization[J]. Knowledge and Data Engineering, IEEE Transactions on, IEEE, 2006, 18(10): 1338–1351.

[19] Zhang M-L, Zhou Z-H. ML-KNN: A lazy learning approach to multi-label learning[J]. Pattern recognition, Elsevier, 2007, 40(7): 2038–2048.

[20] Elisseeff A, Weston J. A kernel method for multi-labelled classification[C]. Advances in neural information processing systems. 2001: 681–687.

[21] Zhang M-L, Zhou Z-H. M3MIML: A maximum margin method for multi-instance multi-label learning[C]. Data Mining, 2008. ICDM’08. Eighth IEEE International Conference on. 2008: 688–697.

[22] Zhang M-L. A k-nearest neighbor based multi-instance multi-label learning algorithm[C]. Tools with Artificial Intelligence (ICTAI), 2010 22nd IEEE International Conference on. 2010, 2: 207–212.

[23] Zha Z-J, Hua X-S, MEI T, et al. Joint multi-label multi-instance learning for image classification[C]. Computer Vision and Pattern Recognition, 2008. CVPR 2008. IEEE Conference on. 2008: 1–8.

[24] Nguyen N. A new svm approach to multi-instance multi-label learning[C]. Data Mining (ICDM), 2010 IEEE 10th International Conference on. 2010: 384–392.

[25] Li Y-X, Ji S, Kumar S, et al. Drosophila gene expression pattern annotation through multi-instance multi-label learning[J]. Computational Biology and Bioinformatics, IEEE/ACM Transactions on, IEEE, 2012, 9(1): 98–112.

[26] 宋相法, 焦李成. 基于稀疏编码和集成学习的多示例多标记图像分类方法[J]. 电子与信息学报, 2013, 35(3): 622–626.

[27] Chen Z, Chi Z, Fu H, et al. Multi-instance multi-label image classification: A neural approach[J]. Neurocomputing, Elsevier, 2013, 99: 298–306.

[28] Zhou Z-H, Zhang M-L, Huang S-J, et al. Multi-instance multi-label learning[J]. Artificial Intelligence, Elsevier, 2012, 176(1): 2291–2320.

[29] Zhang M-L, Wang Z-J. MIMLRBF: RBF neural networks for multi-instance multi-label learning[J]. Neurocomputing, Elsevier, 2009, 72(16): 3951–3956.

[30] Wu J, Huang S, Zhou Z. Genome-wide protein function prediction through multi-instance multi-label learning[J]. Computational Biology and Bioinformatics, IEEE/ACM Transactions on, IEEE, 2014,11(5):891-902.

[31] Yang C, Lozano-perez T. Image database retrieval with multiple-instance learning techniques[C]. Data Engineering, 2000. Proceedings. 16th International Conference on. 2000: 233–243.

[32] Asano T, Chen D Z, Katoh N, et al. Polynomial-time solutions to image segmentation[C]. SODA. 1996, 96: 104–113.

[33] Jiang Y, Chen K-J, Zhou Z-H. SOM based image segmentation[G]. Rough sets, fuzzy sets, data mining, and granular computing. Springer, 2003: 640–643.

[34] Kohonen T. Self-organizing maps[M]. Springer Science & Business Media, 2001, 30.

[35] Zhou Z-H, Zhang M-L, Chen K-J. A novel bag generator for image database retrieval with multi-instance learning techniques[C]. Tools with Artificial Intelligence, 2003. Proceedings. 15th IEEE International Conference on. 2003: 565–569.

[36] Xu X, Frank E. Logistic regression and boosting for labeled bags of instances[G]. Advances in knowledge discovery and data mining. Springer, 2004: 272–281.

[37] Freund Y, Schapire R E, Others. Experiments with a new boosting algorithm[C]. ICML. 1996, 96: 148–156.

[38] Zhang M-L, Zhou Z-H. Multi-instance clustering with applications to multi-instance prediction[J]. Applied Intelligence, Springer, 2009, 31(1): 47–68.

[39] Boutell M R, Luo J, Shen X, et al. Learning multi-label scene classification[J]. Pattern recognition, Elsevier, 2004, 37(9): 1757–1771.

[40] Frank M, Wolfe P. An algorithm for quadratic programming[J]. Naval research logistics quarterly, Wiley Online Library, 1956, 3(1-2): 95–110.

[41] Cristianini N, Shawe-taylor J. An introduction to support vector machines and other kernel-based learning methods[M]. Cambridge university press, 2000.

[42] Wang J, Zucker J-D. Solving multiple-instance problem: A lazy learning approach[J]. Morgan Kaufmann, 2000:1119-1125.

[43] R.e.schapire AND Y. singer. BoosTexter?: A Boosting-based System for Text Categorization[J]. 2000: 135–168.

[44] 王吉权. BP神经网络的理论及其在农业机械化中的应用研究[D]. 沈阳农业大学, 2011.