摘要:
基于高光谱图像的矿物识别主要依赖于矿物的特征光谱曲线,但由于矿物普遍存在的同质多象和类质同象的现象,常常造成矿物识别误差。在矿物的光谱曲线描述中,吸收位置是一个非常重要的特征,它较好地反映了矿物的单个诊断性特征。文章在利用完全波形特征的基础上,加强了单个诊断性特征,提出了一种基于吸收峰加权的预处理技术。利用该技术并结合PMS光谱库(南京地质矿产研究所研制)中的部分数据,从矿物光谱相异性进行区分实验,由于该技术利用了完整的矿物特征信息并突出了重要信息,对矿物有更为全面细致的描述,同时融合了去包络算法,有效地抑制了噪声的影响,与以往主流技术相比提高了矿物识别准确度,拉大了相异矿物间的差异,对研究岩石的内部结构和矿物的种类将有较大贡献。
Abstract:
Hyperspectral image in spectral curve are often used in the field of mineral recognition. However, error recognition may occur because of the polymorphism and isomorphism existing in minerals. The position of absorption peaks play a very important role for the description of spectrum curve,which preferably reflects the single diagnostic feature of mineral. Therefore,Based on using the whole curve's feature, we proposed the preliminary technologies of weighted absorption peaks to enhancing the single diagnostic feature. Using the part of data from the standard spectral library PMS, we has accomplished some experiments based on the dissimilarity of mineral. Because the technology used whole mineral features and highlighted the important features, the description of mineral was more comprehensive and accurate. Meanwhile, the technology applied the continuum removes algorithms to reduce the effects of back noise. Compared with the regular technology, the proposed technology not only improved the accuracy of the mineral identification, but also increased the discrepancy between the different minerals, which will give a greater contribution to study the internal structure and the species of minerals.
下载: