Classification method of noisy label images based on depth learning

Abstract

The invention discloses a classification method of image containing noise label based on depth learning, which relates to the technical field of image classification. The method of the invention comprises the following steps: step 1, performing data extraction on semantic metadata set and image data set to obtain a baseline dataset; step 2, training the baseline model based on the baseline data set, and extracting the characteristic information of the baseline data set through the trained baseline model; step 3, clustering the characteristic information through hierarchical clustering to obtain a new data clas; step 4, re-dividing the baseline data set based on the new data category to obtain the classification data set data; step 5, training a final classification model by using a short_inclusion network base on that data data set; sStep 6, classifying the noisy label images according to the final classification model. The invention solves the problem that the existing classificationmethod of the image containing the noise label cannot be applied to the large data set.

Description

technical field [0001] The invention relates to the technical field of image classification, in particular to a method for classifying images with noise labels based on deep learning. Background technique [0002] Convolutional neural network (CNN) refers to a class of deep learning algorithms for processing data with a known grid-like topology, specifically those networks that use an operation called "convolution", where convolution is a A special form of linear operation, specifically referring to in-place matrix multiplication at more than one network layer. RNN is a deep learning algorithm for processing sequence data. In recent years, recurrent neural network (RNN) has been successfully applied to tasks such as speech recognition, machine translation, and language models. It can be said to be a standard network for processing text data. Hierarchical clustering refers to the successive generation of nested clusters by merging or splitting data sets; there are two main c...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to: in order to solve the problem that the existing classification method of noise-containing label images cannot be applied to large data sets, the present invention provides a classification method of noise-containing label images based on deep learning

Images