Relational reinforcement learning system and method based on graph structure data

Abstract

The invention provides a relational reinforcement learning system and method based on graph structure data, and the system encodes a real world image and an unmatched high-resolution image into degradation representations, and then processes the positive and negative sample relationship between the representations through comparative learning, thereby obtaining an encoder capable of correctly extracting image degradation features. The method is used by a subsequent down-sampling network. A low-resolution image obtained by processing a high-resolution image through a linear down-sampling network has degradation characteristics of a real world, and a finally generated data set is input into a reconstruction network for training to complete an image super-division task. Compared with an existing super-division algorithm based on fuzzy kernel estimation, the method has the advantages that implicit expression between images can be learned without introducing extra degradation models and prior knowledge, and better generalization is achieved while the same performance as a supervised method is achieved.

Description

technical field [0001] The invention relates to the technical field of deep learning and image super-resolution, and more particularly, to a relational reinforcement learning system and method based on graph structure data. Background technique [0002] Image Super Resolution is an important research direction in the field of computer vision in recent years. The purpose of single image super-resolution is to perform end-to-end training on low-resolution images to obtain clearer high-resolution images. Linear interpolation, nearest neighbor interpolation, and bicubic linear interpolation based on sampling theory are emerging in the field of super-resolution, and are a simple and convenient super-resolution method. According to these operators, a high-resolution image can be quickly generated without occupying additional Space. Research in recent years has shown that methods based on convolutional neural networks have achieved more significant results than traditional method...

AI Technical Summary

Problems solved by technology

[0007] In the prior art, a patent of a sampling-based undirected graph structure data comparison method is disclosed. This patent is based on a sampling undirected graph structure data comparison method, and uses random sampling to compare node receiving information to judge two undirected graph structure data. Whether the graphs are the same, compared with the existing graph similarity algorithm, the method designed by this patent has better efficiency, and can be well applied to massive data processing; when the scale of the graph is large, it can also get a lot Good running effect; and when it is judged that the two graphs are different, it can ensure a certain difference, and when it is judged that they are different, the correctness of the result can be guaranteed with a high probability; at the same time, the undirected graph structure data comparison based on sampling provided by the patent The method has good generality and can be extended to the comparison of weighted undirected graphs; however, the patent does not involve any information about training the encoder to extract degenerated features through contrastive learning, and then using the encoder and downsampling network to generate low-level images with degenerated features. High-resolution images, construct pairs of training data, and finally use them again to obtain technical solutions for high-resolution images with perfect details and good visual effects

Images