Research On Change Detection Of Multi-temporal Remote Sensing Images Based On Deep Learning

The task of remote sensing image change detection is to detect the changed area by comparing different time states of the same area,which has important applications in the fields of forest cover and urban planning.With the improvement of remote sensing image quality,it is still a challenge to accurately identify the change regions.In recent years,scholars at home and abroad have made breakthroughs in the research of deep learning algorithms,especially the powerful feature characterization ability of convolutional neural networks makes it outstanding in the image field.Therefore,deep learning algorithms are identified as a way choice for remote sensing image change detection tasks.To effectively identify change regions and further improve the detection accuracy,this paper briefly summarizes the existing remote sensing image change detection methods based on deep learning techniques,and proposes two different remote sensing image change detection algorithms from the twin network framework using U-shaped network structure.The main work of this paper is as follows.(1)In order to effectively explore the real semantic changes,avoid the interference of redundant information,and extract detailed change information while accurately locating the change region,this paper proposes a change detection algorithm based on multi-scale fusion.The algorithm constructs an input pyramid to extract and fuse the feature information of remote sensing images at different scales,which realizes the multi-scale reception of input images,enhances the network’s ability to perceive the image feature information,and avoids the loss of original information to a great extent;the long short-term memory(LSTM)is used to dynamically assign weights to the difference features to discriminate the real semantic change information;a multi-scale calculation method of change features is proposed to aggregate the contextual information of different regions,thus enhancing the network’s sensitivity to feature information and improving the network’s ability to obtain global information of images.(2)In order to fully fuse the edge information of the change region,accurately locate the change region and obtain a complete change detection result,this paper proposes an edge enhancement-based change detection algorithm.The algorithm fits the features of the remote sensing image to be detected by migrating the weights of VGG16(Visual Geometry Group)network to get a more abstract feature representation,which is fully prepared for further processing of feature information;introduces a double attention mechanism,which focuses on the spatial dimension and channel dimension representation of features,integrates different levels of difference features,and gets the real semantic change information.Extract and selectively fuse the edge change information at different levels,optimize the change detection results by the edge information in the change region,and improve the overall detection accuracy;a new hybrid loss function is proposed to better guide the parameters of back propagation during the gradient descent of the network,reduce the influence of change edges,and improve the robustness of the network.