Enhancing Neural Network Efficiency in Automated Image Analysis for Thermal Nondestructive Testing

The article is devoted to the study of the efficiency of using the image fusion method for data augmentation when training neural networks for thermal non-destructive testing.Using the example of training 75 neural networks with the AlexNet architecture, the dependence of image classification accuracy on the data augmentation method was analyzed.The effectiveness of generating artificial images using the method of data integration based on wavelet transformation was examined.The results of the analysis demonstrate the advantages of using image fusion as a method of data augmentation and enrichment to improve the efficiency of neural networks.The superiority of the data augmentation method based on image fusion using wavelet transform is proven.