Deep layer feature fusion-based Chinese traditional visual culture symbol identification method

Abstract

The invention discloses a deep layer feature fusion-based Chinese traditional visual culture symbol identification method, and belongs to the technical field of image processing and computer vision. The method comprises the steps of firstly training a classification model by utilizing a convolutional neural network in deep learning; secondly extracting visual culture symbol features of each layer in the trained model, calculating a weight of each layer by utilizing Softmax regression, and combining the features of each layer into a long vector, which serves as an image feature representation of each type of image; thirdly performing PCA dimension reduction and normalization on the extracted features, and inputting the features to a shallow learning SVM for performing classification; and finally by utilizing an ensemble learning thought, combining an identification result of deep learning with an identification result of deep and shallow learning combination by utilizing a regression tree to obtain a final classification result.

Description

technical field [0001] The invention relates to a recognition method of traditional Chinese visual culture symbols based on deep hierarchical feature fusion, and belongs to the technical field of image processing and computer vision. Background technique [0002] Chinese traditional visual cultural symbols are the historical and cultural heritage of the Chinese nation for thousands of years, and also a symbol of Chinese traditional culture. It condenses the wisdom and strength of the Chinese people, and becomes a way for the world to interpret China. It can represent China and influence the world. The study of Chinese traditional cultural symbols has far-reaching significance. At present, the inevitable trend of globalization has made traditional culture face a major impact. Therefore, it is objective and inevitable to re-examine and inherit Chinese traditional culture, and traditional cultural symbols as traditional culture A kind of symbol and logo, which has a far-reachi...

AI Technical Summary

Problems solved by technology

[0012] (1) In the traditional method, the final feature is sent to the last layer, that is to say, the network will filter out the information it considers useless, but the filtered information is not necessarily useless, and it is also useful for image details. Certain expressive ability, so the shortcoming is that there will be loss of characteristics

[0013] (2) The last layer of the traditional neural network is connected to Softmax, and the Softmax logistic regression model uses the gradient descent method to update the parameters. In this case, the convergence speed will be a bit slow, because there is a process of parameter update

In terms of classification effect, logistic regression is susceptible to interference from outlier data, which affects the accuracy of classification.

[0014] In general, the current pattern recognition-based cultural symbol classification methods can be broadly divided into shallow learning and deep learning methods, but as mentioned above, each type of method has its own shortcomings, especially the cultural symbol recognition based on deep learning. Methodologically little research

Images