Optical excitation infrared nondestructive testing method based on generative adversarial network

Abstract

The invention discloses an optical excitation infrared nondestructive testing method based on a generative adversarial network, and the method comprises the steps: obtaining an original infrared thermal image sequence on a tested piece with a defect through an optical excitation infrared thermal imaging nondestructive testing system, carrying out the preprocessing of the original infrared thermalimage sequence, and forming a training data set and a label data set; establishing and training an improved generative adversarial network and a discrimination network, segmenting defects of two different test pieces in real time, wherein the segmentation effect is better than that of an existing model.

Description

technical field [0001] The invention belongs to the technical field of non-destructive testing, and more specifically relates to a light-excited infrared non-destructive testing method based on a generative confrontation network. Background technique [0002] Non-destructive testing technology refers to the detection of its state characteristics and physical properties without damaging the performance and structure of the measured object. Infrared non-destructive testing technology refers to exciting the object to be tested, recording the change of radiation energy, and judging the physical characteristic information of the object through the inconsistency of the temperature energy change of the defect and non-defect area. Optical excitation infrared non-destructive testing technology refers to the use of light source as the excitation source to quickly and widely stimulate objects to detect defects. In recent years, it has achieved good results in the field of non-destructi...

AI Technical Summary

Problems solved by technology

Existing traditional defect detection algorithms, such as principal component analysis, thermal signal reconstruction, and variational Bayesian methods, are limited by the robustness of the algorithm or cumbersome parameter adjustment. have poor accuracy

So far, end-to-end deep learning defect algorithms have not been widely studied, and most deep learning frameworks have little effect in the field of nondestructive testing.

Images