Zero-sample image classification model based on repeated attention network and method thereof

Abstract

The invention relates to a zero-sample image classification model based on a repeated attention network, and the model comprises a repeated attention network module which is used for training and obtaining the sequence information of an image region; a generative adversarial network module, used for acquiring visual error information; a visual feature extraction network processing module, used forobtaining a one-dimensional visual feature vector of the image; an attribute semantic conversion network module, used for mapping a low-dimensional attribute semantic vector to a high-dimensional feature vector with the same dimension as the visual feature vector by using two linear activation layers; a vision-attribute semantic connection network, used for realizing fusion of a vision feature vector and an attribute semantic feature vector; and a score classification result and reward output module, used for classifying the types with the labels seen by adopting cross entropy loss, and reward output being used for punishing the unseen data without the labels and punishing the prediction result with the highest possibility of the seen types and the unseen types in the data without the labels. According to the invention, the problem of image category label missing can be effectively solved.

Description

technical field [0001] The invention relates to a zero-sample image classification model, in particular to a zero-sample image classification model based on a repeated attention network and a method thereof. Background technique [0002] Currently, in the process of image classification, if you want to accurately classify images, you need to inform the model of the image labels of each category. However, the number of image categories is often very large, and new categories may be added from time to time. If each category is manually labeled each time, the workload will be extremely huge. In this process, some categories have only a few or no training sample labels, and the whole category without training labels belongs to zero samples. Such zero samples cannot be effectively constructed by traditional machine learning methods. The purpose of zero-shot learning image classification is to solve the problem of missing the entire category label and classify the categories that...

AI Technical Summary

Problems solved by technology

However, the number of image categories is often very large, and new categories may be added from time to time. If each category label is manually marked each time, the workload will be extremely huge

In this process, some categories have only a few or no training sample labels, and the entire category without training labels belongs to zero samples. Such zero samples cannot be effectively constructed by using traditional machine learning methods.

Images