Research On Naming And Retrieval Methods Of Cultural Relics Based On Fusion Features

Digitization and intelligence in the field of cultural relics are current hotspots and difficulties in academic research,but the field of cultural relics is highly specialized,and the multi-modal characteristics and rich cultural connotations of cultural relics data set a high threshold for research work,leading to cutting-edge research in the computer field Technology cannot be quickly applied to the field of cultural relics.Through questionnaire surveys of museum visitors in the field of cultural relics and interviews with scholars in the field of cultural relics,it is concluded that the urgent need in the field of cultural relics is to use intelligence to efficiently obtain cultural relics information.This work can be refined into two tasks: intelligent naming of cultural relics and cultural relics retrieval.The intelligent naming of cultural relics is to use cultural relics pictures to generate cultural relics’ names or description information to assist visitors in interpreting cultural relics;cultural relics retrieval is to use the multi-modal characteristics of cultural relics as a similarity measurement standard to improve the relevance of search results and query content.This article focuses on the work of cultural relic information extraction and retrieval.First,build a multi-featured cultural relics data set MFCR for subsequent intelligent naming and cultural relic retrieval.The constructed data set includes two types of cultural relics characteristics of image information and text information.Angle information proposes a word segmentation method based on matching rules and content to extract relevant information;in the research on cultural relics naming,a network structure of Res Net50-LSTM combined with attention mechanism is built,Res Net50 network extracts image features,LSTM network is used for text generation,and attention is added The mechanism allows the model to adjust the image focus during the text generation process,and the role of each part of the network is verified through experiments and the effect of comparison with similar methods on the cultural relics data set is improved;the intelligent naming work is used to generate the cultural relics retrieval work.Descriptive information and cultural relic images are used as fusion features to describe the similarity between cultural relics,and the high-dimensional vector corresponding to the cultural relic features is stored through the product quantization algorithm.The ablation experiment proves that the fusion feature optimizes the recommended results,and compares the cultural relics with other image retrieval methods Relevance of search results on data.