Interactive data extension method based on Poisson image fusion and image stylization

Abstract

The invention discloses an interactive data extension method based on Poisson image fusion and image stylization, which comprises the steps of extracting engine defects by reading annotation information of an original image based on an interactive data enhancement mode of Poisson image editing, and constructing a defect information base; adjusting the training parameters in the style migration model based on a Pytorch framework, and converting the style of the defect information into the style of the corresponding position in the background image to obtain a new defect image; fusing the new defect image and the background image, correcting a fusion boundary by using Poisson fusion, reducing a fusion trace, obtaining an engine hole detection image with a new defect, and using the engine hole detection image with the new defect as an extended image for model training of an engine hole detection image defect detection task; and performing interactive data expansion on the plurality of original images to generate a plurality of new defect images, taking the original images and the new defect images as an image binary classification data set, and judging the new defect images.

Description

technical field [0001] The invention relates to the field of data enhancement, in particular to an interactive data expansion method based on Poisson image fusion and image stylization. Background technique [0002] Existing data extension methods are generally divided into two categories: traditional data extension methods and deep learning-based data extension methods. Traditional data expansion methods include: 1. Flip the image left and right, random rotation, cropping, scaling, translation, etc.; 2. Local deformation adjustment, adding noise disturbance to the image such as Gaussian white noise, salt and pepper noise; 3. Changing the image Color; 4. Change the brightness, contrast, and clarity of the image; 5. Make affine changes to the image, etc. The commonly used methods are cropping and scaling, flipping and color brightness change. The literature [1] summarizes the typical data expansion methods used in the network. In the literature [2], the author mentioned tha...

AI Technical Summary

Problems solved by technology

[0005] The above data expansion methods do provide great help to the training of the model, but they all expand the original image globally, and even when GANs are used to generate brand new data, the output is uncontrollable (for example, if you want a A defect photo of the engine, but it may generate a new photo with no defects or unrealistic defect locations), and for supervised network training GANs that require label information to generate data, label information cannot be generated at the same time, plus GANs The difficulty of training the model further limits the use of this method

Images