Image classification method and device, equipment and storage medium

Abstract

The invention is suitable for the technical field of image processing, and provides an image classification method and system and a storage medium. The image classification method comprises the following steps: extracting a multi-channel feature map of a to-be-classified image; decoupling the multi-channel feature map to obtain a plurality of single-class decoupling feature maps; obtaining an inter-class relationship according to each single-class decoupling feature map obtained by decoupling; for each single-class decoupling feature map, calculating the classification probability of the corresponding single-class decoupling feature map according to the inter-class relationship and the single-class decoupling feature map; and determining the class of the to-be-classified image according to the classification probability of each single-class decoupling feature map. Through feature decoupling and inter-class relation extraction, the existence probabilities of the possibly coexisting features are independently calculated, whether the feature tags exist or not is determined, and the accuracy of image feature classification is improved.

Description

technical field [0001] The present application belongs to the technical field of image processing, and in particular relates to an image classification method, device, equipment and storage medium. Background technique [0002] In image recognition scenarios, several similar features often appear on an image at the same time. When the existing machine learning model learns such samples, it will learn the phenomenon of coexistence between features. Using such a machine learning model to identify images, for multiple features with many coexisting phenomena in the sample, the model usually thinks that when one feature appears, other features will also appear, and it is difficult to identify different features separately, resulting in confusing results, that is affect the classification accuracy. [0003] Taking the recognition of myopic fundus color photos as an example, at present, the recognition of myopic fundus color photos can be realized by training a multi-label classi...

AI Technical Summary

Problems solved by technology

A single sample in the training set may contain one or more classification labels in leopard-like fundus, diffuse atrophy, plaque atrophy and macular atrophy; Dependency, when a large number of categories coexist in the training set, the convolutional neural network will learn a large number of coupled features, making it difficult to distinguish different feature labels separately during reasoning and prediction, and it is easy to confuse label categories with high coexistence frequency

For example, the training set contains many images with leopard-like fundus and plaque-like atrophy coexisting. When the network model infers images with leopard-like fundus, it tends to think that there is plaque atrophy at the same time, which will lead to classification prediction. The results are biased

Images