【作者】 王东；

【导师】 吴湘滨； 毛先成；

【作者基本信息】 中南大学 ， 国土资源信息工程， 2008， 博士

【摘要】 利用现代遥感技术辅助矿产资源勘查,是快速有效的勘查支持方法,但由于遥感影像中各种矿化蚀变信息是一类弱信息,传统方法的提取效果仍存在许多方面问题有待于进一步提高或改进。矿化蚀变信息提取是遥感矿产勘查中的关键技术,因此对现有的技术进行深入研究,将新技术、新方法应用到遥感矿化蚀变信息提取中,提高遥感找矿的效率与可信度,具有非常重要的理论和实际意义。本文结合一项“十一五”国家科技支撑计划项目和一项青海省重大科技攻关项目的研究工作,建立了一种新的基于粒子群智能的遥感找矿辅助方法。通过对大规模组合优化问题的优化和求解方法研究,提出了符合蒙特卡罗算法性质的化简模型,揭示了问题优化解之间的多种性质。将上述研究结论应用于多种智能计算方法的实验结果表明,智能算法的收敛质量得到明显地提高。对比试验结果显示,粒子群智能在离散问题求解中具有较其它智能计算方法更好的搜索性能,确定以粒子群智能作为本项研究工作的核心技术。在对粒子群智能的机理研究基础上,建立了基于粒子群智能的二维离散空间搜索框架模型。模型中二维离散空间中的点赋予引力,粒子在引力作用下以模拟智能生命的概率控制方式飞翔。由于模型中使用了引力机制,使得搜索模型具备常规方法缺乏的全局性,引力衰减机制的引入增强了模型的鲁棒性,因此新提出的搜索模型具有良好的人性化指标。将粒子群智能搜索框架和线性混合像元分解两个模型结合,建立了基于粒子群智能的混合像元分解新方法。首先利用粒子群智能搜索算法进行尝试性分解搜索,然后根据搜索结果再进行线性混合像元分解,在一定程度上解决了常规混合像元分解方法中存在诸如线性配准不可控、误判、缺乏全局性等问题。对比实验结果表明:新分解方法的分解结果更符合影像中目标地物的展布情况,表现了良好的全局性,保留了更多的遥感找矿信息。通过对遥感影像中矿物岩石光谱特征的研究与分析,提出了一种矿物岩石光谱特征“漂移”假说,将该假说与粒子群智能搜索模型结合,建立了基于粒子群智能的矿化蚀变信息提取方法。由于遥感影像分辨率尚没有达到理想状态,所以像元通常是多种地物的混合光谱,造成矿化地物光谱部分波段出现偏移,利用这种现象建立了地物类别划分方法。同时在新方法中使用新的邻域搜索参考模型,增强了方法在搜索过程中对邻域信息的参考强度,使得分类结果的全局性得到了进一步加强。在此基础上,将粒子群智能行为特征以量化方式表示,建立了一种新的分类结果密度分布模型,为下一步遥感找矿奠定基础。针对支持向量机分类器进行遥感矿化蚀变信息的提取,建立了粒子群智能快速优选支持向量机分类器超参数的方法。通过对分类器两个关键参数对分类结果影响情况的分析,确定以整数编码方式以及k-折交叉验证作为适应度评价实现参数优选搜索;同时提出了两点中心法和多点重心法两种启发式策略,进一步提高算法的搜索效率。采用经过优化参数后的支持向量机分类器进行遥感矿化蚀变信息提取,缩短了特征提取时间,分类质量得到了进一步提高。综合上述各种技术建立了基于粒子群智能计算技术的遥感找矿流程,对项目中几个典型矿化区段进行了遥感矿化信息提取应用。通过野外实地验证和与已知的矿(化)区资料对比,提取的蚀变异常信息与已知的矿(化)点位置基本吻合,新发现的一些矿化蚀变异常点均不同程度存在矿化蚀变现象。综合其它找矿资料,给出了4个成矿预测远景区、4处找矿靶区和9处新的找矿线索地段。更多还原

【Abstract】 It is a kind of fast and efficient support way of exploration to utilize modern remote sensing technique to assist mineral resource exploration. There are many issues to be resolved or improved in traditional information extraction approaches because mineralization alteration information is weaker than other information in remote sensing image. Mineralization alteration information extraction is critical technology of remote sensing mineral exploration, so it has important academic and applied significance for improving efficiency and reliability of remote ore-finding to go deep into existing technology, and to apply new technique and new method to remote sensing mineralization alteration information extraction. Research working in this theme, establishing a kind of assist method for remote sensing ore-finding based on particle swarm intelligence computation technique, is based on the 11th Five Years support programs for science and technology development of China and key technologies R&D program of Qinghai Province.Through studying optimization and solving method for large scale combinatorial optimization problem, simplification model is put forward settling for Monte Carlo algorithm property, and manifold characters among optimal solutions are revealed also. The experimental results, applying above-mentioned conclusion to different intelligence computation methods, show that convergence performances of those algorithms are improved obviously. Contrastive experiential results show also that particle swarm intelligence has better search performance than other intelligence computation methods while being applied to solve discrete problems. Therefore, it is treated as kernel technique to be studied in this theme.Searching framework model based on particle warm intelligence is established for two-dimension discrete space on a basis of mechanism study of particle swarm intelligence computation. In the model, points in two-dimension discrete space are assigned gravity to. All of particles fly under gravity action according to probability controlling manner that simulates intelligence life. The new model possesses overall situation which normal methods are absent because it utilizes gravity mechanism. Gravity attenuation mechanism boosts up robustness of model. Consequently, new model is provided with better human indicator.Particle swarm intelligence mixels decomposition method is brought forward through combining searching framework with linear mixels decomposition model. Tentative decomposing search is done firstly utilizing intelligence search algorithm, and then decomposing images utilizing linear mixels decomposition according to search results. This settles some issues existing in general mixels decomposition methods, such as non-controlled property of linear matching, and lack of overall situation and so on. Contrastive experiential results show that decomposition results meet distribution of target landmarks in remote sensing image. New method is with better overall situation, and reserves more remote sensing ore-finding information.Shifting hypothesis of mineralization spectrum character is put forward through analyzing and studying mixture property of mineral spectrum in remote sensing image. Particle swarm intelligence mineralization alteration information extraction method is established through combining the hypothesis with particle swarm intelligence search model. In the meanwhile, new neighborhood-search reference model is added into the new method to enhance ulteriorly overall situation of classification results. On this basis, behavior characteristic of particle swarm intelligence is quantified to established probability distribution model of classification results, and to lay the groundwork for follow-up remote sensing ore-finding.In allusion to remote sensing mineralization alteration information retraction utilizing support vector machine, quick optimization-selecting method using particle swarm intelligence for support vector machine classifier is put forward. Through analyzing influence to classification results produced by two critical parameters of classifier, parameter search method is implemented by way of integer encoding manner and evaluating through k-fold crossover validation. In the meanwhile, two heuristic strategies, two-point-epicenter method and multi-point- barycenter method, are founded for improving search efficiency. All of these reduce time of feature extraction, and improve ulteriorly classification quality. Workflow of remote sensing ore-finding based on particle swarm intelligence is established integrating above-mentioned techniques. New method is applied to several representative mineralization segments of programs. Through field experiments and comparing with data of the known alteration areas, we find that the alteration information is nearly in accordance with the known alteration areas. Alteration points founded newly have alteration phenomenon in some degree. Four metallogenic prediction prospective areas, four mine prospecting target areas and nine new ore-finding clue segments are given through synthesizing other ore-finding materials.更多还原