Research On Remote Sensing Image Change Detection Method Based On Saliency Detection

In recent years,the progress of science and technology and the process of urbanization have continued to advance,and the earth's surface has changed with each passing day.In order to monitor and analyze surface changes in a macroscopic and dynamic manner,multi-temporal remote sensing image change detection has become one of the main technical means.This technology obtains the change information of the ground features before and after multiple observations of remote sensing images and recognizes the characteristics of the surface changes,and then analyzes the changes in the nature,location,and range of various features.Because high spatial resolution remote sensing images contain more complex spatial information,higher requirements are placed on the interpretation capabilities of remote sensing images.As an important part of remote sensing image processing,high spatial resolution remote sensing image change detection has been widely used in the fields of urban expansion,agricultural surveys,environmental monitoring,and disaster analysis.Traditional change detection methods are still slightly inadequate in the processing of spatial characteristics of remote sensing images.Saliency detection has been widely used in the field of computer vision due to its advantages in space and scale.The purpose of saliency detection is to find the foreground target area in the image that can attract the attention of the human eye,and selectively ignore the uninteresting area to make the result more robust.In addition,the detection of salient target regions in challenging and complex scenes has also been effectively verified.Based on the analysis of domestic and foreign related results,this paper applies different saliency detection models to high-resolution remote sensing image change detection to improve the accuracy of change detection.The main contributions of this article are as follows:(1)A remote sensing image change detection method based on clustering Co-saliency detection is proposed.The clustering Co-saliency detection technology not only considers contrast information and spatial information,but also comprehensively considers the correlation information between images.This method first uses Change Vector Analysis(CVA)and Spectral Gradient Difference(SGD)to construct difference images;then performs cluster-based joint significance detection on the difference images and performs significant difference image fusion The Co-saliency map is obtained;finally,the OTSU method is used for threshold segmentation and closed operation processing to obtain the change image.Through the joint significance detection of two difference images,the detection result is the common change area of CVA and SGD.(2)Propose a method to detect regional feature integration and significant remote sensing image changes.In terms of salient features,compared with the existing algorithm to calculate the contrast value of the region,this method calculates a contrast vector and introduces a new feature vector to characterize the background;at the same time,the distinguishing regional feature integration is directly integrated with the saliency learning algorithm The feature vector is used to calculate the saliency map;in addition,the method is based on the region,can perform multi-level estimation,and can capture non-local contrast.The three types of regional features are integrated into one discriminant strategy,and multi-region saliency detection is performed.(3)A method of building change detection based on residual attention network saliency detection in remote sensing images is proposed.This method can be combined with an advanced feedforward network structure in an end-to-end training mode.The network structure is stackable,and the stackable basic module introduces the mechanism of attention feature maps,and different levels of feature maps can capture images.The multiple response results in,use the residual connection method at the same time,so that the attention modules of different layers can be fully learned.Coupled with the advantage of enhanced salient features,the building was extracted as the target in the experiment,and the training process of extracting the building target was strengthened to enhance the classification effect,thereby improving the accuracy of change detection.