A training sample generation method for industrial big data processing

Abstract

The invention belongs to the field of image processing, and specifically discloses a training sample generation method for industrial big data processing. Data set and small sample data set; build an industrial image generative adversarial network and optimize the objective function, and iteratively train the industrial image generative adversarial network based on the optimization objective function to obtain a small-sample generative parameter model; input the large-sample image in the large-sample data set into training The obtained small samples are generated into the parametric model to generate small sample images, so as to complete the generation of training samples. The present invention does not need to perform complex digital image processing operations on the industrial image, and does not need to perform various transformations on the original industrial image, which can avoid excessive manual intervention and reduce errors in industrial image generation caused by the professional quality of operators.

Description

technical field [0001] The invention belongs to the field of image processing, and more particularly, relates to a training sample generation method for industrial big data processing. Background technique [0002] With the introduction of intelligent manufacturing, industrial big data technology has become a key factor in improving the productivity, competitiveness and innovation of manufacturing in the future. However, as the original driving force for industrial big data technology, industrial data, especially industrial image data, still have many problems: 1) The problem of sample imbalance, that is, there is a large gap in the amount of data between various types of industrial image data; 2) The diversity of available industrial image data is insufficient and cannot cover the existing data characteristics of industrial data. This makes models trained with industrial data have low generalization ability and poor robustness. The above shortcomings seriously restrict the...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems of unbalanced samples and poor diversity in industrial image data when using industrial big data technology, the present invention provides a method for generating training samples for industrial big data processing, which is based on the sharing between different categories of industrial image data Features, using confrontational learning to achieve mutual generation between different categories of industrial images to generate a real industrial image dataset, it does not need to model the industrial image, and does not need to transform the original data, the generated industrial image dataset has High image quality, good variety, etc.

Images