Research On Accuracy Analysis And Results Correction Of Land Type Interpretation In Moderate-resolution Remote Sensing Imagebased On Secondary Survey Data

With the continuous development of remote sensing technology,remote sensing technology has become an important means to obtain the status and change of land use in the region.It plays a crucial role in the field of land dynamic monitoring,urban expansion,land use cover,carbon source/carbon sink,ecological environment.Although the spatial and temporal resolution of remote sensing images has been improved to a certain extent andthe high-resolution remote sensing images are widely used in many aspects,but the medium resolution remote sensing images are still researched and applied because of the convenient acquisition and the relatively mature interpretation technology.And the accuracy of the interpretation of the medium resolution image is affected by the spatial resolution and the spectral characteristics of the features,and the classification accuracy is relatively low.The high accuracy of the field survey data due to data confidentiality,acquisition costs and other reasons,can only get some of the regional data,in many applications and research is difficult to obtain a large number of them.This paper takes six counties(cities)in Henan province as the study area,which land use types are interpreted from the moderate-resolution remote sensing imaging represented byLandsat-8 OLI,and the classification method is the support vector machine(SVM)selected from the commonly used basic method through the comparative analysis.The key part of this paper is to use the secondary survey detailed data to analyze the classification accuracy of the land use types of the study area,and try to correct the classification results by building a simple linear model,which purpose is to apply the classification data to estimate the actual data of the main land use types in one area,and facilitate the use of remote sensing technology to classify regional data,with a view to the application of medium-resolution remote sensing image interpretation and correction theoretical and practical basis.The paper is divided into six chapters,specific research mainly includes the following aspects:(1)It takes six counties(cities)of Guangshan County,Yongcheng City,Ye County,Lushan County,Jiyuan City and Lushi County as the research areas based on the principles of value,feasibility,representation and image data integrity.The data of the study area,the elevation data,the remote sensing data and the land survey data are collected,sorted and pre-process,and the data base is provided for the study.(2)The study area will be divided into 6 types,including cultivated area,woodland,grassland,construction land,water and other land,which according to the "Classification of land use status" GBT 21010-2007 and combined with the specific situation of the study area.In this paper,the remote sensing image is extracted by the method of the support vector machine(SVM),and its associated band data supplemented by other data like remote sensing index,slope for multi-source information combination to enrich the remote sensing image data.(3)The paper use the secondary survey detailed data to do a global test of the classification data.Then the classification error data transfer matrix is calculated based on the powerful spatial statistical analysis function of ArcGIS,and the accuracy of classification results is evaluated by calculating the quantitative index of Overall Accuracy,Kappa coefficient.In addition,to analysis the overall accuracy based on the administrative unit of study area administrative village,the statistical analysis of classification accuracy was carried out in different geomorphic units in the study area,and the comparison and analysis of ground Classification Accuracy with the main different land use types.(4)the linear model and the nonlinear model are respectively used to calibrate the classification results based on the correction coefficient method and the Cobb-Douglas production function,and the effect and rationality of the model are tested and verified.The model is used to calibrate the medium resolution remote sensing image Classification results to be corrected,to obtain a more scientific and reasonable land use classification results.Through the above research and analysis,the following conclusions are drawn:(1)The research area of this study is scientific,reasonable and representative,and the support vector machine(SVM)interpretation method is suitable for the land use type classification of the remote sensing image in the study area.(2)Landsat-8 as the representative of the medium resolution remote sensing image of the overall classification accuracy is not high,although the low-resolution remote sensing image land use type classification of the number of high precision,but by confusing the matrix quantitative precision test to see the overall classification low accuracy.(3)The classification of the land types in the medium resolution remote sensing image is affected by terrain and geomorphology and different land use types.In addition,the spatial resolution,characteristics of spectral features,the land cover complexity,etc.also affect the interpretation of the classification accuracy of the land types interpretation.(4)The correction coefficient model established in this study and the correction model based on the Cobb-Douglas production function can better correct the remote sensing image classification results,and the corrected classification results are closer to the baseline data(survey data),which shows that the revised model of the study is feasible.In contrast,the relational formula is established by the correction coefficient,which is suitable for land types such as cultivated land and construction land with small terrain,and the nonlinear model constructed by slope is more suitable for forest type.