Extraction Method And Application Of Winter Wheat From Guanzhong Plain In Random Forest Based On Spatial Characteristics Of Neighborhood Samples

In order to effectively obtain a wide range of winter wheat planting areas in Guanzhong Plain,the Fengyun-3 satellite 250m resolution spectral imager(MERSI)was used as the data source in this paper.Aiming at the problem that traditional sample sampling methods did not fully consider the sample neighborhood structure similarity,combined with crop characteristics of the study area,a sampling model based on neighborhood sample spatial characteristics was constructed.To obtain the samples of the study area,random forest was used to extract the classification map of different phases in the same year,and then voting game theory was used to obtain the crop distribution of different years in Guanzhong Plain from 2011 to 2014.At the same time,driving force was used to analyze the land use situation of Guanzhong Plain in different years.Finally,in order to improve the efficiency of image mapping,a set of random forest classification system based on spatial characteristics of neighborhood samples was developed.Specific work carried out is as follows.(1)Random forest based on spatial features of neighborhood samples for feature extraction in the Guanzhong Plain.Firstly,five to six MERSI images with few clouds and clear images were selected from each year during the winter wheat growing season from 2011-2014,and the data were obtained from the data storage server of the National Satellite Meteorological Center.After pre-processing the data,the region of interest of each image was selected using ENVI5.3,and a neighborhood sample spatial feature model was constructed to obtain training samples and evaluation samples,and the output samples were compared with various classifiers.Finally,the classification data obtained from different simultaneous stacked forests are fused using voting game theory to obtain the final classification result maps,and the accuracy from 2011-2014 reaches 96.53%,91.10%,94.24%,and 98.63%,respectively,while it is basically consistent with the data decoded from LANDSAT8images at the county scale.(2)Analysis of land cover changes and driving forces in the Guanzhong Plain based on random forest classification.Firstly,the change of wheat distribution was analyzed,and the area of winter wheat was 8695.50 km~2in 2011 and 8202.06 km~2in2014,and the area of wheat was changing dynamically from 2011 to 2014,with an overall decrease of 5.67%.Then the transfer of a single type of land was analyzed,urban area increased at a faster rate,mainly transformed by wheat and bare land,wheat wheat decreased mainly transformed into urban and bare land,finally,the land use change from the demographic and economic aspects of the driving force analysis,the growth of population and gross product in varying degrees affect the transformation of other types to urban land,and the economic impact is greater.(3)System design and implementation.Based on the neighborhood sample sampling model proposed in this paper and the random forest classifier used,in order to improve the mapping efficiency,combined with the functional requirements of the experiment,the processing framework is designed.After testing,a system is developed reasonably,including sample sampling module and classification module,so as to effectively improve the mapping efficiency.