Application Research Of Image Recognition Based On Capsule Network

Thanks to the rapid development of artificial intelligence,image recognition technology has penetrated into all aspects of people's lives,how to accurately and efficiently identify various types of images has become a hot spot in the field of computer vision.Deep learning have integrated the feature extraction and classification recognition by simulating the human brain,and have always performed well in the field of computer vision.Capsule network is a new deep learning method.It uses the dynamic routing algorithm between two vectors to replace the pooling layer of the traditional convolutional neural network,which not only avoids the loss of information,but also can well represent the spatial structure of the object to be identified,effectively making up for the shortcomings of the traditional convolutional neural network.However,for specific classification and identification tasks,how to improve and exert its excellent performance remains to be explored.Combining with the advantages of the capsule network,this paper studies the recognition of overlapped characters and rotating image.The specific work is as follows:1.Improved capsule network for recognition of orderly overlapped handwritten numerals.The original capsule network did not consider the order of output characters when recognizing overlapped handwritten numerals.To improve the structure of the capsule network,a recognition model with multiple digital capsule layers was designed to realize orderly overlapped handwritten numerals recognition.The experimental results show that the recognition accuracy of the method reaches87.62%,which is much higher than the convolutional neural network recognition,which proves the effectiveness of the method and also provides a certain reference value for the character-overlapping captcha recognition.2.Character-overlapping captcha recognition based on capsule network.In order to greatly reduce the training cost and realize the ordered recognition of multiple characters by the capsule network,all characters in the image are converted into a vector form,and a captcha recognition model is constructed to make multiplecharacters are regarded as a classification task for recognition.Finally,experiments are performed on the data set to prove the feasibility of the method.3.Remote sensing image classification based on local binary pattern and capsule network.In order to improve the classification remote sensing images accuracy,this paper first performs channel fusion on circular LBP feature images and RGB images,secondly,a 1*1 convolutional layer was introduced to better achieve information integration and enhance the expressive power of the network,In addition,a deeper capsule network structure is designed to enable it to extract higher level remote sensing image information.Tested on the data set and compared with other methods,the experimental results show that the method in this paper not only has good convergence,but also significantly improves the classification remote sensing images accuracy,especially for images with rich texture information,the recognition rate has been improved by more than 4%.