Image single-category classification method and device

Abstract

The invention relates to an image single-category classification method, which is carried out according to the following steps: 1, acquiring a plurality of target images, and putting the target images into a set target image folder; 2, collecting a plurality of non-target images, and putting the non-target images into a set non-target image folder; 3, sending the multiple target images and the multiple non-target images into a selected deep neural network for classification training, and obtaining a classifier after classification training for a period of time; and step 4, inputting an image into the classifier to obtain a classification prediction result of the image. The invention can be used for identification of animals or figures or buildings or objects, can also be used for judgment of product defects, and particularly can be used for automatic inspection and identification of product appearance quality, so that fatigue caused by visual inspection of human eyes is reduced, the detection precision is improved, the inspection working efficiency is improved, and the practical value is very high.

Description

technical field [0001] The present invention relates to the technical field of image classification, in particular to a method and device for single-category classification of images. Background technique [0002] Existing image classification includes the following three methods: [0003] The first type of image classification, for example, to determine whether an image is a cat rather than other objects, you can collect many cat pictures and non-cat pictures, and train a binary classifier to make a judgment, but the classifier trained in this way will be There is a defect of paying too much attention to how to distinguish the positive samples in the training set from the negative samples in the training set, resulting in poor generalization ability. [0004] The second type of image classification is also a single-category training scheme, such as one class SVM (one class support vector machine) and SVDD (support vector domain description), both of which have only one cla...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the invention is to solve the existing image classification, there is too much attention to how to distinguish the positive samples in the training set from the negative samples in the training set, resulting in poor generalization ability, or only the aggregation features of the positive samples are trained, and the positive samples are not considered. The difference between samples and negative samples, or samples are divided into many categories, the problem of low efficiency of image classification

Images