摘要
该项目针对传统融合方法无法正确表征物理信息的缺陷,建立了递变能量X射线成像的物理表征模型。该方法是鉴于神经网络可逼近任意非线性映射的特点,以标准楔形试块为对象,将不同电压下的融合图像作为输入数据,直接采集高动态成像图像作为输出数据,经神经网络训练,构建递变能量成像的物理表征模型。同时在不同种材料下,对物理表征模型进行了修正,实现了不同材质下的灰度校正。利用钢质与铜质阶梯块验证模型。结果表明:该项目提出的算法能逼真地反应直接高动态成像特性,可正确表征工件的物理信息。
The X-ray gradient energy imaging fusion method can not correctly characterize the physical characteristics of detecting objects.So an X-ray imaging physical characteristic algorithm based on variable energy is proposed in this paper.Because the neural network can approximate any nonlinear mapping correctly,the procedure is to take a standard wedge blocks as test objects,and take the fusion images of the low dynamic images as input data and acquire a high dynamic image directly as desired output data.An X-ray imaging physical characteristic model is built by neural network training.For heterogeneous material,the model of physical characteristics are modified.Steel and copper objects are tested using the physical characteristic model.Experiment shows that the result image can reflect the characteristics of high dynamic image,and can represent the structure information of test objects completely.
出处
《核电子学与探测技术》
CAS
CSCD
北大核心
2014年第6期770-774,共5页
Nuclear Electronics & Detection Technology
基金
国家自然基金(61171179
61227003
61301259)
山西省自然科学基金(2012021011-2)
高等学校博士学科点专项科研基金资助课题(20121420110006)
山西省回国留学人员科研资助项目(2013-083)
山西省高等学校优秀创新团队支持计划资助
关键词
递变能量
高动态
物理表征
神经网络
variable energy
high dynamic
physical characterize
neural network
