基于高光谱特征波长选取的不同类型沉积岩识别方法

Identification of different types of sedimentary rocks based on hyperspectral characteristic wavelength selection

为快速无损地实现岩石类型精确识别,以禄丰阿纳恐龙山南缘为研究区,采集3类典型沉积岩样本(泥岩、砂岩和泥灰岩各21块),借助ASD FieldSpec3地物光谱仪获取样本在350~2 500 nm内的高光谱数据,对预处理后的原始光谱进行一阶微分和连续统去除变换,采用马氏距离对全波段光谱进行初步筛选,并使用竞争性自适应重加权算法进一步筛选特征波长,基于全波段和特征波长变量分别建立贝叶斯判别和经过粒子群算法优化的支持向量机识别模型.结果表明,马氏距离结合竞争性自适应重加权算法用来筛选特征波长能提高模型的识别准确率,有效剔除光谱中的冗余信息,其中基于连续统去除光谱构建的组合模型被选为最优沉积岩识别模型,其预测集识别准确率为0.952 4,输入模型的特征波长变量数只占全波段的1.6%.

Rock type identification is a basic part of geological investigation and mineral resource exploration.In order to quickly and accurately identify rock types without damage,three types of typical sedimentary rock samples(21 mudstone,sandstone and limestone) were collected from the south edge of the Lufeng Ana Dinosaur Valley.Hyperspectral data within the range of 350-2 500 nm were obtained with the aid of ASD FieldSpec3 surface feature spectrometer,and the preprocessed original spectra were of first-order differential and continuum removal transformation,Mahalanobis distance was used to preliminarily screen the full band spectra,and competitive adaptive reweighting algorithm used to further screen the characteristic wavelengths.Based on the full band and characteristic wavelength data,Bayesian discrimination and support vector machine recognition models working by means of particle swarm optimization were established respectively.The results showed that Mahalanobis distance combined with competitive adaptive reweighting algorithm could not only improve the recognition accuracy of the models,but also effectively eliminated the redundant information in the spectrum.Of them,the combined model based on the continuum removal spectrum was selected as the optimal sedimentary rock recognition model.Its recognition accuracy was 0.952 4 and the number of characteristic wavelength variables input into the model only accounted for 1.6% of the whole band.