基于DenseNet与声学层析成像的温度场高分辨率重建

High-resolution Reconstruction of Temperature Field Based on DenseNet and Acoustic Tomography

提出一种虚拟观测结合密集连接网络(DenseNet)的两阶段声学层析成像温度场高分辨率重建算法。以总体最小二乘为目标,采用虚拟观测法对超声飞行时间(TOF)进行重建,得到粗网格下的温度分布;然后利用搭建的DenseNet预测细化网格的温度分布信息,采用双输入模型,同时利用模块化输出方法对典型的温度场模型进行数值仿真。结果表明:该算法的重建质量和抗噪性均优于常用的虚拟观测方法、Landweber迭代法、Tikhonov算法及代数重建(ART)方法。

A two-stage high-resolution reconstruction algorithm of acoustic tomography temperature field based on virtual observation combined with densely connected network(DenseNet)was proposed.Aiming at the total least squares,the ultrasonic time-of-flight(TOF)was reconstructed by virtual observation method to obtain the temperature distribution under the coarse grid.Then,DenseNet was built to predict the temperature distribution under the refined grid.Numerical simulation on typical temperature field model was carried out using dual-input model and the modular output method.Results show that the reconstruction quality and noise resistance of the algorithm are superior to commonly used methods,such as virtual observation method,Landweber iteration method,Tikhonov algorithm and algebraic reconstruction technique(ART)method.