| With the acceleration of the urbanization process,the construction land increases greatly.At the same time,due to the influence of uncontrollable factors such as natural disasters,the quantity and quality of available cultivated land resources decrease greatly.In order to ensure food production and maintain social stability,the state began to build well-facilitated farmland.The well-facilitated farmland project has developed rapidly since it was proposed because it can meet the development direction of modern agriculture.The rapid construction of well-facilitated farmland will inevitably cause the change of land use information,so it is necessary to study the classification of well-facilitated farmland.UAV(unmanned aerial vehicle)remote sensing as a new means of earth observation,because of its high efficiency,and the rapidness frequency to obtain high-resolution images of ability has been widely in more and more fields,based on the integration of UAV image acquisition,scale,segmentation and image classification technology through the phases of classification methods,such as well-facilitated farmland of modern agricultural science and technology park of Changshu the classification information of land use,and obtain ideal classification results.Then the method of classification by stages is applied to the classification of well-facilitated farmland in Zhaoba village,Liuhe district,Nanjing city,and good results are obtained.Through the research,the following conclusions are mainly drawn:(1)the ESP(estimation of scale parameters)tool was used to assist in determining the image segmentation scale,and the shape factor and compactness factor parameters were determined in combination with a large number of experiments.Finally,the flower field,lake and aquaculture land were extracted from the segmentation layer with scale=500,shape parameter=0.2 and compactness parameter=0.1,respectively.The bare land was extracted in the segmentation layer with scale=150,shape parameter=0.7 and compactness parameter=0.5,and in the second stage,the bare land was classified into cropless farmland and active land by the nearest neighbor classification at this scale level Drainage channel and shelter forest were extracted in the segmentation layer with scale=50,shape parameter=0.8 and compactness parameter=0.8.Greenhouses,rural roads and houses were extracted in the partition layer with scale=300,shape parameters=0.7 and compactness parameters=0.5.(2)through the phases of the classification method of modem agricultural science and technology park of Changshu well-facilitated farmland classification information of land use,the first stage by the method based on rules of characteristic difference between most obvious feature categories are classified,the second phase by using the nearest neighbor classification will feature the easily confused features of categories are classified,and then through the confusion matrix method to evaluate accuracy of classification result,total classification accuracy is 97.49%,the kappa coefficient was 0.9711,that phase of the classification of the classification method has obtained the good effect.(3)in order to verify the grading efficiency in the information classification of well-facilitated farmland classification method,by using the eCognition in stages classification of software integration process of Liuhe district,Nanjing city Zhaoba village well-facilitated farmland classification,and accuracy evaluation,total classification accuracy is 97.54%,the kappa coefficient was 0.9678,achieved a higher classification accuracy,explain in stages classification method has good applicability in well-facilitated farmland classification. |
