Research On Convolutional Neural Network Based Semantic Segmentation Using Dual Source Remote Sensing Data

Remote sensing image segmentation is a process of dividing images into different regions according to certain similarity criteria.It is one of the basic problems in remote sensing image processing.Image segmentation plays an important role in scene analysis,target detection and 3D reconstruction of remote sensing images.From the development of remote sensing technology to the present,it is still one of the research hotspots and widely used in many fields such as weather prediction,geological exploration and forest fire prevention.However,with the gradual expansion of the application range of remote sensing image data,it is difficult to meet the demand of high resolution remote sensing applications by traditional processing methods for medium and low resolution images.Therefore,it is need to fully understand the characteristics of high resolution remote sensing image data and propose the better analysis method.Through combining with the characteristics of different remote sensing image data and realizing the complementary advantages of different source data,it can improve the ability of information extraction of remote sensing image data.From high resolution remote sensing data and its corresponding elevation data joint application,it adapts the method of transfer learning.Based on convolutional neural network,single source semantic segmentation model of remote sensing data is constructed,and based on this,remote sensing data semantic segmentation model of dual source is proposed.The double source data information processing is completed using the neural network’s strong learning ability to further improve the accuracy of segmentation.The research contents mainly contains the following aspects:Firstly,the development history of deep learning,the basic theory and application of convolution neural network and the typical model of transfer learning are studied.First of all,the development of deep learning is summarized,and then the basic principle and training process of convolution neural network are briefly introduced.Finally,the basic concept of transfer learning and the common network model are introduced.Secondly,based on convolutional neural network,monophyletic semantic segmentation model of remote sensing data is constructed,the migration of learning methods,based on the model of migration,designed and implemented in three different cross layer fusion of semantic structure,segmentation,respectively in high resolution remote sensing data to the semantic segmentation experiment and elevation data,verify the validity of the semantic segmentation model.Finally,the semantic segmentation model of dual source remote sensing data based on total convolution network is proposed.Through the study of the characteristics of high resolution data and elevation data extraction method.first,extract the single source feature and fusion them,then distinct stronger dualsource the fusion feature is obtained.Last,build the convolution network model of dual source to integrate the features,and then put them into the softmax classifier to classify and complete the double source remote sensing data of semantic segmentation,and the method obtains the segmentation result is better than that of single remote sensing data.