| Remote sensing of crops is the key to monitor agricultural condition.Utilizing GF-1 and Landsat 8satellite-remote sensing data in distinguishing cotton,the merits are threefold:Firstly,it supplies agricultural investigation with comparatively accurate and synchronous data;Secondly,it offers precise parameters to the localized application of new-satellite data;Moreover,it provides technical support to saving the cost of cotton area-investigation,effectively achieving the cotton subsidies,and decisions of the government and Agricultural Administrative departments.This research selected Changji as the experimental area,and was based on the remote sensing images collected by GF-1 and Landsat 8.This research used the 2015 field survey of crop planting information,selected different supervised classification methods,and involved mask of arable land,to analyze and obtain the optimal data source of cotton distinguishing,best distinguishing time phase and top identification method.Through analyzing,the following conclusions can be reached:Firstly,The accuracy of outcome involving remote sensing image mask of GF-1 and Landsat 8 all exceeds the the supervision classification outcome without extracting arable land as mask.The application of mask enhances the precision of overall classification.Support vector machine method of GF-1,Landsat-8 improved the overall classification precision by 0.44%,0.79%;the extraction accuracy of cotton planting area has been improved 31.77%,6.38%.Secondly,both the overall classification precision and the cotton user precision of GF-1 remote sensing data surpass those of Landsat 8.The mapping precision of Landsat 8 is better than GF-1.Despite no significant interpreting accuracy variation between the the two data source,GF-1 satellite data brings about comparatively high interpreting accuracy and cotton area extraction accuracy.Support vector machine method of GF-1 interpreting accuracy is 95.24%,the cotton extraction accuracy of it is 95.96%.Thirdly,Using the Mahyagra distance,minimum distance,parallel hexahedron of supervised classification method to supervise and classify the remote sensing image data of GF-1 and Landsat 8,on the base of support vector machine and with method of maximum likelihood to determine the best recognition month and optimal recognition approach,the result of this research shows that:July has the most ideal time phase of recognizing the cotton area in both these two data source.In the interpretation accuracy of the GF-1 image data is 95.24%,kappa coefficient is 0.935;the counterpart of Landsat 8 is93.89%and 0.916.Support vector machine method has been the Supervise classification method with the highest interpretation accuracy so far.The interpretation accuracy of maximum likelihood method ranks second.In terms of operating rate of the two classification approaches,the maximum likelihood method performs better.Fourthly,When extracting area,the cotton recognition information interpreted by support vector machine method utilizing the images collected by GF-1,it results nicely when the minimum map spot is less than six pixel.On account of this scheme,the cotton area interpreted is 15974 hm~2 in Changji2015,survey area is 16647 hm~2,the extraction precision of cotton area is 95.96%. |
