Change Detection Of High Resolution Imagery Based On Machine Learning And Fuzzy Fusion

Remote sensing change detection plays an important role in many application scenarios,especially in the monitoring of geographical conditions,which is closely related to our living environment.However,there is no general method that can obtain perfect results for the change detection of different scenes,and the existing change detection methods have their own limitations due to their own background.Based on high resolution remote sensing image change detection,this paper combines the feature of high resolution remote sensing images of high image characteristics and differences to study the multi-scale features and timely difference.It is found that due to the high resolution remote sensing image features of the scale of the diversity and the characteristics of object oriented algorithm,change detection results on different scales are difficult to be effectively integrated.While objectoriented change detection on single scale is often difficult to obtain reliable results and object feature misses the feature edge information to some extent.In order to solve these problems,this paper proposes a method of multi-scale fusion based on object features and machine learning,which considers pixel –based fusion in object-feature change detection.The key point is to put forward a general fusion method to help realize the abstraction and standardization of the change detection process.Firstly,this paper describes the method selection and reasons of pretreatment,image segmentation,detector and reliability evaluation involved in the multi-scale change detection process,and verifies them by theoretical description and specific experiments.Then,based on the spectrum and texture features,the change discrimination of single scale fuzzy evaluation is carried out by support vector machine(SVM).It is found that compared with the common method of using membership function for change probability discrimination,the change detection effect of using posterior probabilistic SVM for change discrimination is better.Finally,under the premise of object feature optimization and optimal scales,the acquisition and fusion of change detection results based on object features and multiple scales are analyzed emphatically,and the effective fusion of change area guided by different scale change probabilities is realized through entropy weight thought in information theory.In this process,in order to improve the degree of automation of the change detection algorithm,the single-scale object-level features are assigned to the internal pixels of the object to facilitate the automatic processing of the image.The results show that the proposed method can not only reduce the uncertainty of single scale change detection results,but also reduce the dependence on segmentation scale in traditional object-oriented change detection.