Deep ECT Image Enhancement Method Based on Mixed Precision Training

Abstract

The invention discloses a depth ECT image enhancement method based on mixed precision training, which belongs to the field of image processing. Firstly, Gaussian noise is randomly added to the preliminary reconstructed image, which is divided into training set and test set. The neural network is then trained using the training set. The reconstructed images of the test set are respectively input into the trained encoder network, and the image features are automatically extracted by using the basic feature extraction network; then through the hollow space pyramid network, the corresponding outputs are connected in the channel dimension of the feature, and after 1×1 volume After the convolution, the image features extracted by the basic feature extraction network after 1×1 convolution are connected in the channel dimension, and finally the output of the encoder network is obtained. Finally, the output of the encoder network is input into the decoder network, and the size and feature information of the image are restored through the combined operation of 1*1 convolution and upsampling by 2 times and the multi-level fusion operation, and the reconstructed image is output. The invention avoids loss of image information and improves image reconstruction speed.

Description

technical field [0001] The invention belongs to the field of image processing, in particular to a deep ECT (electrical capacitance tomography) image enhancement method based on mixed precision training. Background technique [0002] In the field of ECT, traditional image reconstruction algorithms include iterative algorithms and non-iterative algorithms. Non-iterative algorithms are fast, but the accuracy of reconstructed images is not high and the resolution is low; iterative algorithms are slower than non-iterative algorithms, but the accuracy of reconstructed images is higher. . [0003] Compared with the non-iterative algorithm, although the traditional iterative algorithm can improve the reconstructed image quality to a certain extent, the overall imaging quality is still low and the resolution is not high. [0004] With the successful application of neural networks in the image field, many researchers use neural network technology for ECT image reconstruction. [000...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the problem that the existing technology cannot identify multiple (>2) dielectrics in the multiphase flow distribution and the image quality is not high during the reconstruction process, the present invention proposes a deep ECT image enhancement method based on mixed precision training. Under the premise of improving the imaging quality, the calculation speed is accelerated and the time required for imaging is shortened

Images