Application Research Of Deep Learning In Image Detection And Recognition

The rapid development of digital publishing rely on a complete online sales system.For paper books or electronic publications,a good online display and sales platform can bring convenience to users and improve user experience.At present,major Internet sales platforms still have big problems.In online books,there is a lack of display of specific information on the book cover,copyright page,book catalog information and back cover page.The information displayed on the book details page does not correspond to the products currently on sale also occurs from time to time,and some manually uploaded pictures have the problem of picture misalignment,occlusion,and blurring.Most book covers and back covers are still displayed manually,which seriously hinders the process of books being put on the shelves,is not conducive to the development of enterprises,and greatly affects the user experience.To solve such problems,image classification ability is necessary,to realize the recognition of the front and back cover images of books,and to integrate and display the book copyright pages and related chapter information well.This paper conducts in-depth research on this classification task,collects a large number of book cover and back cover image data sets,and proposes two sets of solutions to realize the classification task for book cover and back cover,and respectively introduce corresponding improvement schemes.One scheme is based on the K-nearest neighbor algorithm,and the other scheme is based on the CNN model.For the KNN algorithm,we designed 7 sets of experimental schemes with different K values and introduced an edge detection algorithm to optimize the experiment,which further reduced the time complexity and improved the classification rate.For CNN algorithm,we built 9 sets of CNN models with different network structures to achieve the front and back cover classification tasks.In the optimization plan,we introduced a data augmentation method to expand the data set to the original 13 times,effectively improving the classification accuracy and generalization ability of the CNN model.The realization and improvement of the two sets of solutions provide an effective theoretical and practical reference for the classification of book covers and back covers.