Research On Spatial Information-Based Remote Sensing Image Change Detection In Posterior Probability Space

As a technical means for real-time dynamic monitoring of global land cover types,change detection change detection is of great significance to related fields.although high-resolution remote sensing images can provide more detailed ground object spectrum and spatial information.Therefore,it is difficult to detect changes based on high-resolution images only based on spectral information.Determine whether a pixel has changed.Recent research shows that spatial image information should also be considered for high-resolution remote sensing images while spectral information is considered.In recent years,a lot of research work has introduced spatial information in change detection based on random field models,such as Conditional Random Field(CRF).Among them,CRF,due to its flexibly settable potential energy function,can introduce spatial information in various ways to improve change detection accuracy.Most of the current CRF-based change detection methods mainly use the change intensity map generated by Change Vector Analysis(CVA)as the observation field.However,the change magnitudes of different change types in the CVA change intensity map are different,which has a negative impact on the accuracy of CRF-based change detection.Since the change intensity map generated by the Change Vector Analysis in Posterior Probability Space(CVAPS)method has the same range of change ranges for different change types,combining CVAPS and CRF will likely improve change detection accuracy.In addition,another critical factor affecting the accuracy of remote sensing change detection is the noise pollution in remote sensing images,such as salt and pepper noise,Gaussian noise,and the mixed noise of the two.Noise pollution will further increase the complexity of the spectral information of remote sensing images,affecting the accurate estimation of the change vector in posterior probability space and decreasing change detection accuracy.Therefore,introducing the anti-noise algorithm in the CVAPS change detection framework can alleviate the noise interference to a certain extent,ensure the estimation accuracy of the change vector in posterior probability space,and further improve the anti-noise capability of CVAPS change detection.Aiming at how the change detection algorithm can effectively use the spatial information of high-resolution images to improve the detection accuracy,this paper couples the Hybrid Conditional Random Field(HCRF)with the FCM-SBN-CVAPS algorithm to improve change detection accuracy in high-resolution remote sensing imagery.On this basis,the anti-noise capability of the change detection algorithm is improved by using the spatial FCM algorithm.The research content and results of the thesis are as follows:(1)Research and propose a hybrid conditional random field(FCM-SBN-CVAPS-HCRF)change detection method based on fuzzy C-means clustering and a simple Bayesian network.For the rich spatial information of high-resolution remote sensing images,CRF can flexibly introduce spatial context information to reduce the adverse effects of complex ground object structures and high spectral variability on change detection performance.Among them,while HCRF introduces the spatial information of the neighborhood,it also takes into account the influence of object-level spatial information on changing pixels.However,both the observation field and the label field of HCRF are built on the basis of the changing intensity map generated by CVA.However,in the CVA change intensity map,the change ranges of different change types are different,and it is difficult to guarantee the change detection accuracy of the HCRF model.Aiming at the problem of inconsistency in the range of spectral changes,the FCM-SBNCVAPS algorithm,as a change detection algorithm based on the CVAPS framework,can model the phenomenon of the same object with different spectra and different objects with the same spectrum in images through FCM and SBN,and accurately estimate the posterior probability space change vector of the multi-temporal image,and judge whether the pixel changes according to the change intensity.Since different change types of posterior probability change vectors have the same magnitude,the change intensity map in posterior probability space is more suitable as the observation field and label field of HCRF,which can better guarantee the change detection accuracy of the HCRF model.The experimental results show that the FCMSBN-CVAPS-HCRF change detection algorithm can better consider the spectral information and spatial information of high-resolution remote sensing images,and reduce the same object with different spectra,the same spectrum with different objects,complex ground object structures and high spectral variability that adversely affect the change detection performance.Compared with CVA-MRF and CVA-HCRF,its change detection accuracy is higher.(2)Research on anti-noise change detection of medium and low-resolution remote sensing images based on spatial fuzzy C-means clustering and simple Bayesian network(spatial FCMSBN-CVAPS).For low-and medium-resolution remote sensing images,FCM is sensitive to noise and cannot effectively decompose mixed pixels,so FCM-SBN-CVAPS cannot accurately estimate the change vector in posterior probability space.In this thesis,five spatial fuzzy Cmeans clustering algorithms(FCM＿S1,FCM＿S2,KFCM＿S1,KFCM＿S2,and FILCM)that can effectively decompose mixed pixels under noise pollution conditions are combined with Simple Bayesian Network(SBN)respectively.Under the framework of CVAPS,five remote sensing change detection methods that can resist Gaussian,salt and pepper,and mixed noises are implemented.The comparative experiments prove that the spatial FCM-SBN-CVAPS algorithm has good noise resistance to Gaussian,salt and pepper,and mixed noise in the medium and low-resolution remote sensing images.(3)A hybrid conditional random field(spatial FCM-SBN-CVAPS-HCRF)algorithm based on spatial fuzzy C-means clustering and a simple Bayesian network for anti-noise change detection in high-resolution remote sensing images was studied.For the problem of noise pollution in high-resolution remote sensing images,this paper combines spatial FCM-SBN-CVAPS with HCRF.Spatial FCM-SBN-CVAPS establishes the change intensity map in the posterior probability space to mitigate noise interference.At the same time,HCRF can further reduce noise interference by introducing spatial information.The experimental results showed that the spatial FCM-SBN-CVAPS-HCRF(FCM＿S1-SBN-CVAPS-HCRF,FCM＿S2-SBN-CVAPSHCRF,KFCM＿S1-SBN-CVAPS-HCRF,KFCM＿S2-SBN-CVAPS-HCRF,and FLICM-SBNCVAPS-HCRF HCRF)has good resistance to Gaussian,salt and pepper and mixed noise in high-resolution remote sensing images.