Accuracy Assessment Methods Based On Spatial Sampling For Remote Sensing Classification

The successful launch of high IV in 2016 has realized a high and stable integration of remote sensing satellite structure, the development of remote sensing technology enables increasingly higher resolution of remote sensing images. As a kind of data products, the amount of data and its update rate are increasing in high speed. The effective methods of the remote sensing image classification accuracy assessment become one of research hot spot. Being a simple and efficient survey method, sampling method has been applied to the accuracy test of remote sensing image classification. However, with the influence of sample size and space sample of the traditional sampling method, there are some problems such as low efficiency and information redundancy. Concerning for the characteristics of wide range of information, multi-temporal, multi-spectral of remote sensing images, developing an effective method guarantees the quality and use of remote sensing images.Applying the spatial sampling theory to the accuracy assessment, this paper proposed an efficient accuracy assessment method for remote sensing image classification. First of all, Spatial correlation analysis on the remote sensing image classification, by Moran’s I spatial model based on remote sensing image as the relationship between the distance and the correlation between the elements to quantify, according to quantify the results and the degree of clustering analysis of remote sensing data to understand the various types of land remote sensing image space distribution. Secondly, according to calculations based on Moran’s I spatial analysis algorithm to determine the spatial fabric swatches programs to address the problem space swatches in remote sensing image data of the sample point distribution. By combining spatial swatches method based on Moran’s I with accuracy assessment error matrix method based on remote sensing image classification were calculated, solves the problem of low accuracy in remote sensing image classification. Finally the rationality of the method is validated with a regional real data as well as its feasibility. The main contents of the paper are summarized as follows:(1) Introduction of background and significance of the researchThis paper analyzed the necessity of the remote sensing image classification accuracy assessment. Summarized the advantages and disadvantages of spatial sampling theory in remote sensing image classification accuracy test, and difficulties existed in the research of accuracy test of remote sensing images.(2) Introduction of theoretical basis and background in remote sensing image classification accuracy assessmentBy analyzing the characteristics of remote sensing data, the necessity of the use of remote sensing image classification accuracy assessment in space swatches methods was proposed. The study also explained the definition of spatial sampling theory referred to herein and the accuracy of testing methods.(3) An accuracy assessment model of remote sensing image based on Moran’s IUsing Moran’s I to analyze the relationship between the distances among pixels and there relevance, computing the optimal sample size and distribution method, and by using the error matrix, the efficiency of accuracy assessment method is valuated.(4) Experiment and AnalysisUsing the remote sensing image of Shanghai and Wuhan, comparing with traditional sampling method, the proposed model of remote sensing accuracy assessment is proved to be valid. The result shows that: based on Moran’s I, the proposed method shared the same accuracy with full accuracy assessment, meanwhile, reduces the sample redundancy effectively and improved accuracy test efficiently.