Research On Content-based Identification And Retrieval Methods Of Ethnic Patterns

In order to give full play to the tacit knowledge of the interpretation and transformation of works in the digitization of intangible cultural heritage of national patterns,and solve the problem of quickly finding the target patterns in the context of big data,this study proposes a content-based method of national pattern image recognition and retrieval,mainly through the establishment of semantic annotation model,to achieve the transformation of low-level features to high-level semantics In order to realize the batik pattern retrieval,the following four aspects are studied: mapping and establishing the batik pattern data set,improving the depth convolution network model to form multi task learning,and designing the batik pattern retrieval system:1.Batik pattern is a special pattern with rich semantic and historical connotation.This paper analyzes and studies the semantic characteristics of batik patterns,establishes a reasonable semantic label annotation model,and gives consideration to professionalism and collaboration,so as to provide an effective basis for semantic annotation of batik patterns.2.In order to overcome the problems of unclear semantic annotation structure,poor structure and information redundancy caused by manual annotation under a large amount of data,irrelevant items are added on the basis of dictionary learning,and manual annotation is replaced by automatic annotation to reduce the workload and improve the quality of annotation.On the basis of this research,batik pattern data set batik dataset is established by processing the collected batik patterns,which is ready for the follow-up research.3.Aiming at the feature of multi-dimensional information of batik pattern,a batik pattern retrieval method based on multi task learning is proposed.This method uses the best pre training model to extract text features.Based on resnet50,the image features are extracted through the improvement of network structure,and two loss functions are defined to realize pattern retrieval of multiple prediction tasks.The effectiveness of this method is verified on batik dataset.4.According to the above-mentioned batik pattern retrieval method,a batik pattern retrieval system is designed,which not only realizes the rapid retrieval of batik patterns,but also provides users with batik knowledge integration services.Through the above work,a research idea and method system of content-based ethnic pattern recognition and retrieval method is finally formed.