Application of Machine Learning Methods for Material Classification with Multi-energy X-Ray Transmission Images

摘要

Automatic material classification is very useful for threat detection with X-ray screening technology. In this paper, we propose the use of machine learning methods to the problem of fine-grained classification of organic matters based on multi-energy transmission images, which has been overlooked by existing methods. The method which we propose consists three main steps: spectrum analysis, feature selection and supervised classification. We show detailed analysis of the relationship between feature dimension and material classification accuracy. Our method can also be used to find optimal X-ray configurations for material classification. We compare the performance of several machine learning models for the fine-grained classification task. For the task of classifying three categories of organic matters, we can obtain the classification accuracy higher than 85% with only X-ray measurements with the dimension of four. In conclusion, the results of our paper provide one promising direction for the automatic identification of organic contraband using multi-energy X-ray imaging techniques.