A Lightweight Regression Network Construction Method Based on Prior Filtering

Abstract

The invention discloses a method for constructing a lightweight regression network based on a priori filtering. A designated degradation operation is performed on each original image in an original image set, a corresponding degraded image is obtained, the original image and the corresponding degraded image are cut into image blocks, and training is obtained. Sample pairs; perform clustering in the training sample pairs, and divide the training sample pairs into different categories according to the clustering results; Ternary quantization: A lightweight regression network is constructed and trained by using the prior filter after ternary quantization. After the training of the lightweight regression network is completed, the input degraded image can be reconstructed end-to-end to a higher quality image. The invention uses a small number of fixed a priori filters to build a network with strong pertinence but lighter weight, simple calculation, fast training speed, small required storage space, and is more suitable for small sample problems.

Description

technical field [0001] The invention relates to the field of deep learning, in particular to a method for constructing a lightweight network based on a priori filter. Background technique [0002] Deep learning is the most promising direction in the field of machine learning. It is a hierarchical machine learning method including multi-level nonlinear transformation. By constructing a network structure suitable for different tasks, it has achieved far better results than traditional algorithms. In computer vision Breakthroughs have been made in areas such as object detection, natural language processing, speech recognition, and speech analysis. With the deepening of research, while the depth and size of the deep network model are approaching the accuracy limit of computer vision tasks, its depth and size are also increasing exponentially, and its required computing consumption and hardware costs are also increasing, which limits the depth. The network is widely used on mobi...

AI Technical Summary

Problems solved by technology

Recently, a fixed-parameter filter network Local Binary Convolution Neural Networks (LBCnn) based on LBP ideas uses LBP-based filters, but this network uses a large number of stacked filters to improve recognition accuracy, and the parameters are still massive.

Images