Zero-shot image classification based on factor space

Abstract

Image recognition technology has become a popular research topic with the progress and development of big data and artificial intelligence technology. This study applies factor space to the semantic embedding space, maintains the consistency between the high-level semantic and low-level image feature spaces, and establishes a direct connection between the data features and information expressed by images to bridge the semantic gap in zero-shot learning. The factor space of the improved algorithm was applied to the image classification model to improve the traditional convolution neural network model and build a new zero-shot image classification model using the AWA2 experiment dataset. A zero-shot was proposed and compared with the traditional zero-shot image classification method DAP and IAP. The new algorithm cannot only reduce the operation time but also improve the zero-shot image classification performance.