Image Reconstruction for Ultrasonic Tomography using Multi-channel Convolutional Neural Network
【摘要】：As an important part of Process Tomography(PT), Ultrasound Tomography(UT) has the advantages of no invasion, no radiation, and low cost. However, UT systems are usually physically limited by the transmitter's emission angle of the transducer, and only three to five receiving transducers which are opposite the excitation transducer can receive the signal. This limitation will lead to insufficient prior information, insufficient data utilization, and severe reconstructed image artifacts, which will affect its practical application. In order to improve the accuracy of reconstruction, a multi-channel Convolutional Neural Network(CNN) method is proposed. The main structure of the U-net network, input of the network has two channels: channel one is the image reconstructed by traditional algorithm, and channel two is the measured voltage value. The training database contains 23,355 samples. After 235 epochs of training, the correlation coefficient of verification set is 0.85 and compared with LBP and SART, the proposed model has better reconstruction results.