基于免疫蚁群优化模型的多源遥感湿地精细分类知识发现

Because of the complex environmental gradients (highly spatial variability) and the similar background characteristics (humid underlying surface) of wetland, remote sensing (hereafter RS) image classification of wetland at plant community scale must meet the following conditions: (1) utilize high spatial resolution RS data; (2) employ multisource data to improve information dimension of wetland observation; (3) explore novel classification methods which can effectively integrate high dimension multisource data. Considering the previous issues, this project designs the following study contents: firstly, select Honghe National Nature Reserve (with plenty of wetland plant communities) as study area, and employ high spatial resolution multisource data (including multi-platform optical RS, microwave RS, terrain information and GIS ancillary data) as data sources; secondly, introduce current research focus of spatial data mining field -- artificial ant colony method to RS data classification and information extraction field, and introduce immune mechanisms to control and management component of artificial ant colony system to establish an artificial immune inspired ant colony system (AIACS) with powerful data mining ability; finally, utilize AIACS to search the optimal feature subset from high dimension data through iterative experiment test and accuracy verification, so as to effectively mining wetland features of spectral, radiation, scattering, texture and spatial geographical distribution information which are stored in multisource data, and a theory and technique system that can integrate multisource data for accurate classification and recognition of complex object (wetland) will be constructed based on artificial immune inspired ant colony optimization model. The implementation of this project will enrich wetland RS classification method and have very important scientific significance and practical application value.

湿地复杂的环境梯度（空间变异性高）和相似的背景信息（下垫面湿润），决定了实现植被群落尺度的湿地遥感精细分类须满足以下条件：使用高空间分辨率遥感数据；综合使用多源数据，增加湿地观测信息维；开拓新颖的能兼容高维多源数据的信息处理方法。 因此，本研究拟选择湿地植被群落类型丰富的洪河自然保护区为实验区，以高分辨率多源数据（包括多平台光学遥感、微波遥感、地形信息以及GIS辅助数据等）为基础，将空间数据挖掘领域热点的人工蚁群方法引入遥感数据分类与信息提取中，并在蚁群的控制管理部分引入免疫控制机制，构建人工免疫蚁群系统，通过反复实验测试与精度验证，从高维数据中遴选最优属性组合，充分挖掘多源数据所蕴藏的湿地光谱、辐射、散射、纹理及空间地理分布特征，构建基于免疫蚁群优化模型的、能兼容多源数据的复杂地物遥感精细分类识别理论方法与技术体系。本研究丰富了湿地遥感分类方法，将具有重要的科学意义与实际应用价值。

湿地复杂的环境梯度（空间变异性高）和相似的背景信息（下垫面湿润），决定了必须引入先进的技术方法实现高精度智能化湿地遥感分类。本研究将数据挖掘领域热点的人工蚁群优化和免疫智能引入复杂湿地遥感分类与信息提取，研究了人工蚁群系统的基本控制机制及多态蚁群的分工及表达方法，初步建立了人工蚁群优化模型并实现了高维数据属性降维与约减；研究人工免疫系统的选择、克隆、变异及克隆抑制等控制机制，建立面向遥感分类任务的人工免疫智能算法；将遥感分类规则挖掘转换为蚂蚁路径寻优问题，初步实现了面向复杂湿地地区的智能化遥感分类规则挖掘技术方法；引入人工智能领域先进的群集智能，建立了基于群集智能的遥感影像非监督分类方法。.本项目基本实现了预定研究目标，包括发展了高维数据降维及分类规则抽取的智能化技术方法，构建了基于免疫蚁群智能的湿地遥感分类技术体系，拓展了群智能算法在复杂湿地遥感影像分类中的应用。未来研究可从以下方面继续加强：（1）在降维与分类规则抽取研究中，设定的目标函数均为单一目标，而实际应用中可能有多种因素影响最优解集的搜索，因此未来应引入多目标优化理论构建更符合实际应用的数据降维与分类规则抽取方法（2）现有智能算法在多源数据处理方面仍然存在巨大困难，未来应该探索更为先进的算法，进一步提高多源数据的智能化处理程度。