Study On Object-based High-resolution Image Change Detection Based On Land Cover Conversion Logic

Land cover change detection has important scientific research and application value.Remote sensing technology provides important data support for the study of land cover classification and its changes.Especially with the development of high-resolution remote sensing satellite imagery,remote sensing change detection has become an important source of land cover change data.However,in the process of change detection after high-resolution image land cover classification,the land cover classification error will cause serious error accumulation effect,which makes the accuracy of land cover change detection become lower.At the same time,there are often unreasonable situations in the results of land cover type conversion.These illogical conversion results affect the accuracy of image change detection.These results often require manual processing in general change detection studies,which greatly increases the time and labor costs of remote sensing change detection.This paper considers these illogical results and other inaccurate land cover conversion results due to errors into two parts to solve separately.One of them include image acquisition conditions,data noise and other errors generated during the Image processing.And another part of the error generated during the process of detecting changes using images mainly includes errors caused by two or more images during registration,errors caused by image classification or target extraction,and error accumulation of two or more images.Therefore,aiming at these two aspects of error,this paper proposes to improve the accuracy of high-resolution image land cover change detection and the automation level of data analysis by using object-based image analysis method combined with land cover conversion logic to improve Bayesian soft fusion method.(1)The object-based image analysis method can not only use a variety of features in the classification process to improve the classification effect,but also effectively reduce the noise in the data acquisition process by using image object analysis.At the same time,studies have shown that object-based image change detection has a better mitigation effect on image registration error problems that are easily generated during high-resolution image change detection.Based on the analysis of object image,this paper discusses how to avoid the error in this aspect.(2)By comparing the classification and comparing with Bayesian soft fusion method,and comparing the effects of different direct change detection methods in Bayesian soft fusion.It is proved that the Bayesian soft fusion method can effectively improve the detection accuracy of changes.At the same time,how to better implement Bayesian soft fusion method is discussed.(3)However,the Bayesian soft fusion method can improve the accuracy and reduce the error in the post-classification comparison,but it still inevitably leads to illogical results,which seriously affect the automation level of land cover change.On the basis of summarizing the main driving factors of land cover change,this paper analyzes the causes of various land cover categories,and proposes a general model for the analysis of land cover causes to facilitate the construction of general land cover conversion.Logic,and it uses the spatio-temporal data model to express land cover conversion logic in mathematical form.In order to finally introduce the land cover conversion logic into the change detection process of the Bayesian soft fusion method,these unreliable land cover conversion results are automatically eliminated.(4)At the same time,this paper will use the Bayesian method based on object and land cover conversion logic to apply to the study of land cover change in the built-up area of Lanzhou and its southeastern margins using two-stage domestic high score image data.The effect of Bayesian soft fusion method on reducing the cumulative effect of land cover change classification error is verified.It proves that the application of land cover logic can effectively eliminate the illogical change detection results,so that the total accuracy of land cover change detection results reaches 95.06%.