Image super-resolution reconstruction method based on fused attention mechanism residual network

Abstract

The invention provides an image super-resolution reconstruction method based on a fused attention mechanism residual network, and solves the problems of poor image reconstruction quality and non-idealvisual effect in the prior art. The researched image reconstruction method comprises the steps of S1, performing data acquisition and preprocessing, and obtaining a training image data set and a to-be-reconstructed image data set; S2, building a network model, wherein the model structure comprises a feature extraction layer, a feature learning layer and an image reconstruction layer; S3, initializing, training and storing model parameters to obtain an optimal model structure and an optimal parameter set; and S4, performing image super-resolution reconstruction, inputting a to-be-reconstructedimage, and outputting a high-resolution image under a corresponding amplification scale. The image super-resolution network provided by the invention combines global and local residual structures, fuses channel attention and spatial attention mechanisms, pays more attention to high-frequency information of the image, reserves important features in the image to the greatest extent, reduces repeated and redundant features, and greatly improves the details and definition of the reconstructed image.

Description

technical field [0001] The invention relates to an image super-resolution reconstruction method, in particular to an image super-resolution reconstruction method based on a fusion attention mechanism residual network, and belongs to the technical field of image processing. Background technique [0002] Image resolution indicates the amount of information stored in an image, usually expressed as "horizontal pixel value × vertical pixel value", and is an important indicator to measure image quality. Generally, the higher the resolution of an image, the richer the details it contains, the more information it provides, and the better the image quality. Therefore, images with higher resolution have important application value and importance in various computer vision tasks. Research prospects. However, due to cost issues, it is usually unavoidable to be subject to conditional restrictions or other noise interference during image acquisition, storage, and transmission in reality,...

AI Technical Summary

Problems solved by technology

Including dictionary-based learning methods, using sparse coding, but limited by the linear representation ability of dictionary learning, the reconstruction effect is limited

In recent years, with the application of convolutional neural network (CNN) in the field of computer vision, representative network models such as SRCNN have emerged in super-resolution reconstruction methods, but these reconstruction methods also have some problems, such as the lack of Attention to high-frequency details, slow convergence, long training time, etc.

[0007] After searching, it is found that the Chinese patent application number 201811439902X discloses a method for super-resolution reconstruction of remote sensing images based on channel attention convolutional neural network. Although the network model of this method is also composed of feature extraction layer, feature learning layer, image Reconstruction layer composition, but its feature extraction layer only uses one layer of convolution operation to extract shallow features, only the channel attention mechanism is introduced in the residual module structure of the feature learning layer, and the overall connection structure draws on the recursive network structure, but there is no Ensure that the parameters of each module are shared, and the effect of recursive training is not achieved. The upsampling operation of the image reconstruction layer adopts a transposed convolution (deconvolution) method, and additionally introduces a 0-filling operation, which will amplify noise interference; the application number is 2019101492716 A Chinese patent application discloses an image super-resolution reconstruction method based on a convolutional neural network. The network model of this method is also composed of a feature extraction layer, a feature learning layer, and an image reconstruction layer. In the residual module structure of the feature learning layer, only The attention mechanism is introduced, and the upsampling operation is used in the decoding stage of the U-shaped module to restore the image to the target resolution size, which will greatly increase the amount of parameters in the feature learning stage. The overall connection structure draws on the dense connection structure of DenseNet, resulting in training There are too many parameters in the medium, and the operation cost and overhead are very high. The image reconstruction layer only uses one layer of convolution operation to restore high-resolution images.

Images