Abstract: There are noises and pseudo-pixels in the ultrasonic tomography of concrete due to the ill-posed nature of inversion problems and the errors introduced from the testing and computation.To reliably identify the defects from the ultrasonic tomography of concrete by enhancing the SNR of inversion images，there are two methods，two-dimensional median filtering and cluster analysis based on SOFM neural network，are used in the post-processing of inversion images.The example of application shows that the combined usage of median filtering and cluster analysis may suppress the noises in the local and connectivity areas and non-local non-connectivity areas.Meanwhile，it is also shown that the solution of the inversion image is greatly improved and the defects are detected more correctly and reliably.