Research On Image Segmentation Method Based Of Improved Capsule Network

At present,with the continuous development and improvement of computer technology and hardware equipment,people’s requirements for image processing and analysis are becoming higher and higher.As an important branch in the field of digital image processing,image semantic segmentation has constantly updated the demand scene and is more strict about the accuracy of segmentation results.In recent years,with the vigorous development of deep learning,some people proposed using convolutional neural network to realize image semantic segmentation,and achieved good results.However,due to the limitations of convolutional neural network,the segmentation details are often not ideal.Therefore,we choose to use capsule network to complete the task of image semantic segmentation.Capsule network can extract the spatial features between objects and the detailed features of the edge part.In the image segmentation task,it can better identify different objects and segment more clearly.The main innovations and achievements of this paper are as follows:(1)In view of the large amount of parameters of the capsule network,which is not conducive to training,a local EM dynamic routing algorithm and a method of sharing parameters are proposed,which can significantly reduce the amount of parameters of the capsule network without reducing the accuracy of the capsule network.(2)A classification capsule Network Inc-EMcap with local EM dynamic routing algorithm is proposed.In the EM matrix capsule network,the concept module is added to extract the information of different dimensions at the same layer,and then iterate through local dynamic routing.Compared with the traditional capsule network,Inc-EMcap network must not only have less parameters,but also have higher accuracy.(3)A semantic segmentation scheme of capsule network image based on u-em matrix is proposed.The U-shaped structure is used to realize the step-by-step connection of the network,and the DCA feature fusion algorithm is used in the feature fusion,so as to increase the correlation of similar features and reduce redundancy.In the experiment,it is verified that the u-em matrix capsule network can segment the picture better.