Flower Recognition Research Based On Wide Residual Network And Migration Learning

The traditional flower recognition algorithm mainly extracts features by hand,and finally uses the classifier to train.Its generalization ability has certain limitations and the accuracy is difficult to break through.The concept of "deep learning" has attracted the attention of many domestic and foreign researchers.Especially in the agricultural field,the use of deep learning technology to obtain agricultural information and guide agricultural production has been widely used.At present,the deep learning network has been able to expand the network depth to thousands of layers,but when the number of layers reaches thousands of layers,the cost per cent% increase in accuracy is huge.Therefore,it has been proposed to reduce the network depth and increase the width of the network,that is,the wide residual network,and achieve good results.At the same time,using deep learning technology research,the research results can not be well displayed on the PC side and the mobile phone side,and the core concept of migration learning is that even the accuracy of the shallow network model is close to the accuracy of the deep network model,and it is better.It is APPlied to the PC side and the mobile end.Through experimental comparison,the combination of migration learning and wide residual network is used to give advantage of both.Using less layers and sample size,the best performance network model is obtained,and this model is APPlied to the recognition of flower images.The main contents are as follows:1.Research and analysis of the status quo and significance of flower recognition technology,and in-depth discussion on migration learning methods and wide residual network;2.Design comparison experiments verify the superiority of migration learning between different depths of the same network model;3.Design comparison experiments verify the superiority of migration learning between different network models;4.Perform performance tests on a flower recognition model based on a migration learning method combined with a wide residual network;5.Design and implement the flower recognition APP.The study found that using migration learning can make the accuracy of shallower wide residual networks close to the accuracy of deeper wide residual networks.The trained model not only has an advantage in the accuracy of recognition,but also can be used on the PC side and the mobile terminal side.The wide residual network based on migration learning is APPlied to flower recognition.The results show that the wide residual network and the migration learning flower recognition model have higher accuracy than other flower recognition models.Combined with the theory of this paper,a simple flower identification software for campus of Shanxi Agricultural University was designed and developed,which realized the recognition of the flower of the campus of Nongda University and achieved good results.