Fruit image freshness attribute migration method based on adversarial network

Abstract

The invention relates to a fruit image freshness attribute migration method based on an adversarial network, and belongs to the field of computer vision in the computer discipline. The method comprises the following implementation steps: constructing a training set, and dividing fruit images into four types according to different freshness degrees, including fresh surfaces, dark surface colors, wrinkled surfaces and partial rotted surfaces; preprocessing image files, including removing error files, removing repeated files, unifying file formats, unifying image sizes, numbering the image files,and grouping the image files according to appearances; constructing a generator network and a discriminator network of fruit images with different freshness, wherein the model prototypes of the generator network and the discriminator network are cyclic consistency generative adversarial networks; separately using the image training sets with different freshness types as a model input for trainingto obtain four different generator models; and taking the fruit image of which the freshness needs to be changed as the input of the model to generate a fruit image corresponding to the freshness.

Description

technical field [0001] The invention belongs to the field of computer vision under the computer discipline, and in particular relates to a freshness attribute migration method based on an adversarial network for unpaired fruit images. Background technique [0002] The freshness of fruit is one of the important bases for fruit pricing. The conventional evaluation of fruit freshness mainly relies on sensory recognition, but this kind of method requires a lot of manpower and material resources, and the evaluation standards are different; it can pass physical and chemical tests in the testing laboratory. Or non-destructive testing and other methods for accurate evaluation, but such methods require professional equipment and cannot be widely used. [0003] The development of computer vision technology enables computer vision technology to quickly classify and identify fruits. Using image processing technology based on computer vision to automatically identify and classify the fre...

AI Technical Summary

Problems solved by technology

[0004] Image style transfer is a method of extracting the features of two images and transferring the features of one image to another image. The features are fused and generated based on the style of the second image, with the first image as the original image. New image technology; but the freshness attribute is a kind of fuzzy attribute description, so the freshness attribute migration should be a mapping between image sets, rather than attribute migration between single samples, and the image style transfer technology is directly applied to fruit images The transfer effect of the freshness attribute is not good, and even generates "four different" fruit image results

Images