Water Change Detection Based On Pixel-level Fusion Of Optical And SAR Images

The change detection of surface information is closely related to our lives.The production and construction of human society,the normal succession of natural environment,the outbreak of and other disasters,such as floods and earthquakes,will cause changes in surface information.And the detection of these changes can provide important theoretical and decision support for our life construction and our work to combat the sudden disasters.Remote sensing technology,because of the convenience of large-scale surface observation,provides great convenience for surface change detection.At present,with the rapid development of satellite technology and the increasing of sensor platforms,the acquisition quality of remote sensing images has been optimized,and the coexistence of multi-types and multi-resolution of remote sensing images has also been created.Optical sensor image and synthetic aperture radar,as the most important two kinds of sensor data,have their own advantages and disadvantages.Although perform well in change detection,they are less able to cooperate with each other and do not make full use of the existing surface information.According to this problem,this paper uses the efforts of pixel-level fusion of entinel-1A SAR and entinel-2A MSI imgaes to detect the change of surface information based on Unet.By comparing the efforts with that of non-fusion image based on threshold method,we can determine the contribution rate of image fusion and Unet,and and analyses the relationship between fusion processing and the final detection accuracy.The main research contents are as follows:(1)Image fusion.There are many kinds of pixel-level image fusion methods.In order to take into account all kinds of methods as much as possible,this paper chooses some representative methods to perform fusion processing from three aspects of component substitution,multi-scale transformation and model,and chooses root mean square error and mutual information as the two evaluation indicators from image spectral fidelity and information enrichment respectively to evaluate the Accuracy of all the fusion methods used.(2)Detection based on Unet.The role of Unet in image in information extraction has been proved many times.In this paper,we use labelme tool to make samples of the efforts of fusion image obtained by different fusion methods,and carry out training and detection work to obtain the final detection results.Finally,we get the detection accuracy through statistical methods to evaluate the detection accuracy.(3)Analysis of the role of pixel-level image fusion and Unet in water change detection.In order to determine the specific significance of image fusion processing and Unet in the practice of land surface change detection,this paper compares the detection results of fusion images besed on Unet and non-fusion images based on threshold,and obtains the contribution rate of Unet and the image fusion methods used in this paper to determine their roles in change detection quantitatively.In addition,the overall relationship between fusion accuracy and final detection accuracy is roughly determined by drawing the fusion accuracy-detection accuracy change curve.Finally,it is found that there is a positive correlation between fusion accuracy and change detection accuracy.(4)Water change detection in Ningxiang city.According to the analysis results of the roles of image fusion and Unet in detection,the optimal image fusion method was selected and combined with Unet to detect the water changes in June 2017.This thesis has 31 figures,13 tables and 106 references.