基于Ada Boost的核素识别方法

Methodology of Nuclide Identification Based on AdaBoost Boosting Algorithm

论文介绍了在真实能谱衰减环境中,提出一套对核素进行识别学习算法的流程。采用SVD特征抽取对能谱数据降维,提取到能谱特征向量,在形成的特征向量数据集上,训练决策树分类器,进而通过AdaBoost集成学习算法对多轮的决策树算法的训练结果进行融合,使用K轮类别投票法结合策略,构建一个结果更为接近标签值的假设函数算法模型,解决了探测器检测的能量信息具有局部特征、存在重叠峰值导致核素判别出现错失误判的问题,提高核素识别率。

This paper introduces a nuclide identification method based on AdaBoost algorithm in an environment of truly energy spectrum decay. Reduce dimensionality of the nuclides energy spectrum datasets by SVD,then obtain an feature vectors datasets,Apply Decision Tree Classifier algorithm on the feature vectors datasets to output training result. Using weighted combination strategy of different Decision Tree Model’s outputs based on AdaBoost alogorithm,build a high accracy function model in which its result approximate equals to ture label value. The model boost the weak perfomance of singal decision tree algorithm,Finally,build a model which have lower error rate and more effective generalization performance,identify the nulclide label exactly,this solve the problem about lower rate of nuclide identification in the traditional identified territory.