基于人工智能的异常地物光谱自适应剔除及分类算法研究

Artificial intelligence based algorithm for using spectrum to adaptively eliminate exceptional data and automatically classify

针对传统光谱数据预处理与分析的现状,提出一种基于人工智能的光谱异常数据自适应剔除及自动分类算法,通过遗传算法的优化搜索确定马氏距离的阈值,实现异常光谱的自适应剔除,并提出可量化光谱剔除效果的异常一致性指数(ACI)。在此基础上,借助自组织神经网络方法,以各类观测对象的特征光谱作为输入对象,对剔除后的光谱进行自动分类。经过实验验证,算法取得了较好的剔除效果(ACI达到86%以上)和分类效果(总体分类精度达到94%),较好地实现了异常光谱剔除和光谱分类的自动化处理。

The spectral data measured from spectral measurements are easily affected by human,environmental,equipment and other factors leading to the abnormal spectral characteristics and impacting analyses especially in spectral measurements in the fields.According to the situations of traditional spectral data in preprocessing and analysis,a novel algorithm used for abnormal data excluding adaptively and spectral data classifying automatically based on artificial intelligence was established.The Mahalanobis distance threshold by genetic algorithm searching was determined to exclude abnormal spectral data adaptively and to quantify the effect of excluding abnormal spectral consistency index(ACI).With the self-organizing neural network,spectral characteristics of various types of observing objects were used as input and classified automatically after removing the abnormal.The results showed that the algorithm achieved good excluding(ACI more than 86%)and classification(overall classification accuracy of 94%).It can be used to well automate the handling of excluding spectrum and spectral classification.