High Resolution Remote Sensing Image Research On Intelligent Change Detection Technology

Remote Sensing Image Change Detection is a process of analyzing and determining the changes of various features over time from remote sensing images acquired at different times for the same region.Since remote sensing image data can be obtained in full-time and wide-area,it plays an important role in urban construction,land to use,disaster prevention,land equity confirmation,road network design,crop to yield estimation,environmental resource monitoring,and national defense construction.It can provide important decision support for economic and social development and national security.Due to the existence of the “Spectral Aliasing” effect,the difference between the change feature and the unchanging feature is getting smaller and smaller.It is more and more difficult to determine the change region and the unchanging region by the traditional threshold segmentation algorithm.It is very difficult to improve the accuracy of change detection.Therefore,how to make full use of remote sensing image feature information to explore the detection method with better robustness and accuracy is an important research direction in the field of remote sensing image change detection.The research focus of this paper is on feature fusion of change detection,threshold segmentation,depth feature extraction and detection,and focus on the key areas of the feature map.This paper mainly completed the following three aspects of work:1.Aiming at the problem that the traditional change detection methods can only extract a single feature,the difference intensity map is sensitive to the threshold segmentation,and the traditional threshold segmentation algorithm does not fully utilize the feature information,this paper proposes a change detection algorithm based on the fusion multi-feature and the Two-Level Clustering,In order to make full use of the feature information,the threshold segmentation precision is improved,thereby improving the detection effect of the change.2.For the traditional change detection method,the change information is obtained through the underlying features of the image,and different data sources have different expression capabilities for various features.In the process of change detection,a large number of experimental analysis are performed on the data source,in order to overcome the above difficulties,This paper proposes Pseudo-Siamese Network with multi-level training method extracts high-level abstract features for a change detection,making the expression of change information more complete and robust.3.For the general Convolutional Neural Networks training samples,the slices are used for training,it has the problem of large amount of data,high data redundancy,and large computational cost.This paper proposes Deep Siamese Convolutional Network for the change detection to solve this problem.The input and output data are processed by the whole scene image,and a large number of slices are omitted as training data and test data,so that the quantity redundancy is greatly reduced,and Dual Attention Network is introduced,so that the Convolutional Neural Networks has the ability to focus on its feature subset.