Research On Remote Sensing Image Object Level Land Cover Change Detection Method

Land cover change detection is not only the main content of global environmental change research,but also the key basis for making decision for sustainable resource utilization.Using remote sensing image for land cover change detection is the most convenient and economical method to obtain land cover change information.Therefore,the research on land cover change detection method based on remote sensing image is an important aspect in the field of remote sensing application.Combining remote sensing technology,statistical methods and machine learning,this paper proposes two methods for detecting land cover change,and validates the effectiveness of the method by using GF-1 and GF-6 satellite images at different times in the same region.The specific research contents are as follows:(1)Research on the method for determining the best features of change detectionIn order to solve the problem that the computational complexity increases as the feature dimension increases,it is necessary to study the feature optimization method.The paper uses the analysis of variance to reduce the feature dimension from 25-dimensional to 12-dimensional in the detection of land cover change based on GF-1 satellite imagery.The feature dimension is reduced from 25 to 8 dimensions based on GF-6 and GF-1.The feature dimension is reduced,the calculation efficiency is improved,and the basis for subsequent land cover change detection is laid.(2)Study on the method of determining the threshold of double constraint condition changeFor the traditional change detection method,on the basis of a single constraint condition,the change threshold is determined by human subjective,so that the change detection result is not objective,object-level double constraints land cover change detection method under is proposed,which uses the change vector analysis and correlation coefficient.At the same time,Kappa coefficient is used as the evaluation index to determine the optimal change threshold.Finally,objective and accurate land cover change detection results can be obtained.(3)Research on automatic selection method of best effective samplesUnder a small number of samples,for achieving higher accuracy of land cover change detection results and reducing the amount of time and effort,the paper proposes an object-level land cover change detection method based on active learning double constraints,which solves the problem of time-consuming and laborious manual labeling,and realizes more intelligent and objective and accurate land cover change detection.(4)Comparison test and verification of different change detection methods under double constraint conditionsIn order to obtain the applicability and effectiveness of the above methods,the paper uses GF-1 and GF-6 images to carry out experiments,and compares with the traditional object-level change vector analysis land cover change detection method in terms of detection results and detection accuracy.In the GF-1 image-based experiment,the object-level land cover change detection method based on the active learning double constraint condition and the object-level land cover change detection method with double constraint conditions increased the total detection accuracy by 13% and 16%,respectively;In the GF-6 image-based experiment,the object-level land cover change detection method based on the active learning double constraint condition and the different change detection method under the double constraint condition increased the total detection accuracy by 12% and 16%,respectively.The validity and accuracy of the two methods are verified.The object-level land cover change detection method based on the active learning double constraint condition has significant improvement in accuracy.In addition,the innovations of this article are as follows:(1)It is proposed to use the statistical significance test method to optimize the features and reduce the computational complexity;(2)Combining active learning with double constraints,the method of land cover change detection based on active learning double constraints is proposed,which reduces the time and effort spent on manually labeling samples,and obtains more objective and accurate change detection results..This paper improves the efficiency of change detection by using feature optimization,double constraint conditions and active learning to detect land cover change,and enhances the objectivity,accuracy and intelligence of change test results.Based on this results will provide more accurate data for national land planning,land resource data update,and land resource management.Also it will provide technical reference for land cover change applications based on remote sensing images.