Recognition Of National Patterns Based On Deep Learning

National culture is the accumulation and essence of civilization in a nation’s long course of development,and it is irreplaceable and irreparable.National patterns are the essence of national culture and have strong national characteristics.National patterns can reflect the humanities and spiritual features of various nationalities at different historical stages.However,at this stage,due to the serious decline of national culture,members of many nationalities are unable to identify and distinguish their national patterns well.Therefore,correct classification of ethnic patterns is also an effective means of protecting ethnic culture.Because the contents of ethnic patterns are complex,difficult to distinguish,and large in number,if they are manually labeled,it requires a lot of manpower and material resources,and the accuracy rate cannot be guaranteed.Deep learning is a research hotspot at this stage,and it has achieved very good results in many areas of image and speech.It can learn and extract relevant features from a large number of data samples to infinitely approximate the optimal solution.Therefore,this paper proposes a national patterns recognition algorithm based on deep learning.The steps of the specific algorithm are as follows: The data set is preprocessed first,and the deep learning network is used to extract the features of the data set.Then the features are fused.Finally,KNN and related distance functions are used to complete the classification.According to the characteristics of the ethnic pattern data set,the following improvements have been made in image preprocessing,network structure,and feature fusion: In terms of preprocessing,the data set is enhanced to prepare for training of deep learning;Three improved networks and two traditional algorithms are used to extract features of ethnic pattern datasets.Deep learning networks are responsible for extracting more advanced semantic features,and traditional algorithms are responsible for extracting more basic texture contour features;Fusion of the features extracted by different methods can achieve the purpose of complementing advantages and improving accuracy;Dimension reduction is performed on the fused features to remove redundant features to reduce running time.The experimental results show that the algorithm proposed in this paper issuperior to other classic algorithms,and the accuracy rate can reach 98.5% on the test set,and the running time of the proposed algorithm is lower than the time of other comparison algorithms in this experiment.