Research On A Modified Remote Sensing Image Change Detection Based On RNN

Change detection refers to the use of multi-temporal remote sensing images covering the same surface area to determine and analyze the changes of ground objects,which is widely used in the fields of land use survey,resource dynamic monitoring,post-disaster analysis and assessment,and urban planning and construction.With the establishment of multi-resolution,omnidirectional and all-weather earth observation network,the amount of remote sensing data increases sharply,and higher expectations are set for the accuracy and efficiency of change detection.Most of the change detection methods ignore the time relation between remote sensing images and fail to fully exploit the information of time dimension,resulting in low accuracy and insufficient generalization.Recurrent Neural Network can mine the information of time series data.Based on this characteristic,scholars introduce Recurrent Neural Network into remote sensing change detection and analysis,and the accuracy of change detection has been improved.However,the current applications of recurrent neural networks in change detection are mostly based on spectral features,failing to make full use of the various features provided by remote sensing images,and the accuracy of change detection is limited.Based on the study of remote sensing image features and cyclic neural network,an improved detection method based on recurrent neural network is proposed.This paper focuses on remote sensing image features and recurrent neural network to carry out a series of studies on the issues involved in remote sensing image change detection.The contents and innovations of this paper are as follows:(1)Modify the input of the recurrent neural network,and build a multi-feature fusion change detection model based on the recurrent neural network.n this paper,we use a variety of pixel level features of remote sensing image,such as spectral features,texture features,remote sensing index features,and extract their spatial features by calculating the mean value through window operation,input the above features into the cyclic neural network model for multi feature fusion,extract pixel-space-time feature,and realize the extraction of remote sensing image change information.(2)The validity and applicability of the method is verified using remote sensing image data of various resolutions.In this paper,landsat-8 satellite image and sentinel-2a satellite image were used for change detection,and the types of ground object changes were analyzed.The experimental results show that the method presented in this paper has high accuracy and good generalization ability,and can deal with the change detection task of different medium resolution images and meet the practical application requirements.