Research On Zero-shot Image Classification Technology Based On Factor Space

With the progress and development of big data and artificial intelligence technology,image recognition technology has become a hot issue in research.In traditional image recognition tasks,the types of objects that appear in the training phase and the test phase of deep learning are the same,but whenever a new type of object appears,it can only be retrained to recognize it,which increases the cost of the recognition system.In order to solve this problem,the concept of Zero-shot Learning has emerged,and this concept has become a research hotspot for image recognition and classification.In zero-shot image classification,the trained image samples do not include all object categories.Knowledge is transferred by using attributes as a shared knowledge space between known object categories and unknown object categories.When new object categories appear,new object types can still be identified.This article first introduces Zero-shot Learning.Zero-shot Learning can complete the prediction of category labels of unknown samples through the knowledge space shared between known and unknown categories.Aiming at the semantic gap problem in Zero-shot Learning,apply factor space to semantic embedding space,so that the highlevel semantic space is consistent with the underlying image feature space,and the data features are directly related to the information expressed by the image.The conjunction and reduction of factors,expansion and closure of factor space are studied according to the relationship between semantics.The existing factor space algorithm is improved,and the factor space algorithm is improved and designed based on machine learning related algorithms,making the new algorithm more suitable for zero-shot image classification technology.Aiming at the problems of low attribute prediction accuracy and low recognition accuracy of the existing zero-shot image classification model,the traditional convolutional neural network model was improved,and a new zero-shot image classification model was built,and the new algorithm was applied to the model to improve the performance of the network model.Using the AWA2 dataset for experiments,the improved factor space algorithm is applied to the zero-shot image classification model proposed in this paper.Compared with using the traditional zero-shot image classification method,the new algorithm can not only reduce the calculation time,but also improve the performance of zero-shot image classification.