Flat-Field Correction of X-Ray Tomographic Images Using Deep Convolutional Neural Networks

It is proposed that neural networks be used to solve the problem of flat-field correction. A process is described for selecting parameters of a deep convolutional neural network in order to solve the problem of flat-field correction with the instability of an empty beam, training this network, and checking its operability on the generated data. The procedure is tested on data obtained with laboratory X-ray and synchrotron sources.