Application Of Feature Selection And Incremental Learning Of Neighborhood Rough Sets In Image Classification

With the continuous development of science and technology,we have entered the "artificial intelligence" era.Research in the field of artificial intelligence includes speech recognition,computer vision,natural language processing,and robotics.Image classification as an important research topic of computer vision is widely used in intelligent transportation,target recognition and image retrieval.For massive images brought about by the rapid development of smart devices and multimedia technologies,how to design a classification model with high accuracy and high efficiency at the same time is still a challenging issue.At present,people have mastered the image classification technology has made rapid progress,but there are two issues in the image classification is more and more prominent.First,because of the complexity and diversity of the image content,it is necessary to express the image content through multiple features fusion,which will bring a large number of redundant features to reduce the accuracy and efficiency of image classification.Second,most image classification methods are only applicable to still image data sets.For dynamically added image data sets,the original classifier can no longer meet the classification requirements for new image data sets,and retraining classifiers will have a lot of time cost and space memory consumption.In order to overcome the above problems,the feature selection and incremental learning methods of the neighborhood rough set are applied to image classification.This paper main contents and innovations are as follows:(1)Use more complementary SURF and HOG features for images;Expands the related concept of the neighborhood rough set and maps the image data into the neighborhood rough set model;Combining spatial pyramid matching model,this paper proposes an image classification method based on feature selection of neighborhood rough sets;On the Caltech101,Caltech256 and Corel-1000 data sets,the effect of the method on the size and classification accuracy of the visual dictionary is analyzed through experiments.The appropriate feature selection parameters are selected and compared with different special feature selection algorithms to verify accuracy and efficiency of the classification method.(2)Combined with the content of incremental learning of neighborhood rough sets,this paper proposes an image classification method based on conditional entropy incremental learning.This method updates the classifier through the knowledge of the historical learning of the original image dataset,and improves the efficiency of the classifier while ensuring the accuracy of image classification.Analysis the cases of whether there is a new image category in the newly added image.The accuracy and validity of the method are verified in the Corel-1000 data set.