| With the continuous development of society and economy,the research on the accurate extraction of the spatial distribution of the planting structure in the irrigation area has become increasingly important for the refined management of agricultural water and the estimation of crop yields,and its accuracy has a direct impact on agricultural water management.Remote sensing technology has obvious advantages in the field of rapid crop information acquisition.This article is based on high-and medium-resolution multi-source remote sensing satellite data(Sentinel-1,Sentinel-2,Landsat 8),based on high-precision field survey data,using remote sensing data fusion methods,combined with object-oriented crop classification,to solve crops The problem of low classification accuracy.According to the research goal,design the experiment:1)Planting structure extraction based on multi-source remote sensing data.Combining spectral data,radar data,texture information,and classification parameters,object-oriented methods are used to extract crops in the study area.2)Comparison of extraction accuracy of single source data and multi-source data planting structure.Use single Sentinel-2 multi-spectral data for object-oriented single-scale segmentation and multi-scale segmentation,and compare the accuracy of multi-source remote sensing number multi-scale classification in planting structure extraction.The main conclusions of this article are as follows:1)The 8 texture factors obtained by texture extraction through the gray level co-occurrence matrix GLCM,not every texture factor added is helpful to improve the accuracy of feature classification.The texture factor at different window scales has different sensitivity to different features.The separability is calculated by JM distance.The best texture scale for distinguishing houses is 13*13,and the best texture scale for distinguishing rice is 15*15.The best texture scale of woodland is 17*17.Through the nearest neighbor classifier of eCoginition Developer 8.9,the best texture combination is explored as HOM and VAR.The texture combination under the best window scale can effectively improve the accuracy of ground object classification of remote sensing images.2)In the study area,based on the multi-scale segmentation algorithm,first set the scale factor,shape factor,and compactness to 175,0.7,and 0.6,respectively,which can well express the contour of the rice field.After the rice field block classification is completed,Set the scale factor,shape factor,and compactness to 246,0.6,and 0.5,respectively,and the outlines of corn and woodland can be clearly described for classification and extraction.3)The addition of radar data eliminates the interference of cloud and water vapor,and significantly improves the accuracy of ground object extraction,which improves the accuracy of water area extraction by 17.27%.4)The overall classification accuracy of the classification results combined with multi-source remote sensing information feature combination is 86.64%,the Kappa coefficient is 0.8268,the overall accuracy of many spectral classification results is improved by about 7.55%,and the overall accuracy of the classification results combined with single-split scale multi-spectral data is increased to 9.13%.The classification results combined with the multi-source remote sensing feature combination in housing roads,woodland,rice and water areas.The production accuracy of the classification results is improved by 5.72%,1.38%,13.28%and 14.97%compared with the Sentinel-2 multi-spectral data,indicating that in multi-source remote sensing Image classification of rice fields,corn fields,and forests has good applicability.Figure[34]Table[15]Reference[81]... |
PDF Full Text Request