| Farmland information is the important basic data in realizing"precision agriculture". The important position information of farmland information derived form field boundary. Flied information extraction is very important to realizing regional "precision agriculture" and improving the accuracy of the cultivated land. Sichuan province has complex geomorphic types, whose field distribution difference is big in different area. Especially the field coefficient. Its width lies between 0.3 to 0.8 m, which can influence the cultivated field (plot) boundary extraction and acreage of cultivated land. Therefore, this article choose Jinling village, TianMa town in dujiangyan city of Plain region in ChengDu, hilly area in basin of Longxu village, Qi long town, FuShun county, mountain area around the SiChuan basin of Guang Hui village, GuLin town, GuLin county as the research area.use four kinds of detecting methods of different area field boundary extraction are discussed, Object-oriented technology, panchromatic HP detection, panchromatic Sobel detection and panchromatic Roberts on ENVI4.8 software platform. Determining suitable extraction method in the study area according to the classification accuracy and Kappa coefficient provides technical support for the similar area when it needs the field boundary extraction research. The main results are shown as follows:(1)The best combination band is 431 according to the study of the band combination of Pleiades-1 remote sensing image in the 3 study areas. The optimal band index of the plain region of ChengDu, hilly areas of basin and mountain areas around the SiChuan basin was 646.8,179.93 and 646.8, which maximize keep the spectral information and reduces the interference between each band. Respectively used the HSV transformation, Brovey transformation, "Gramm-Schmidt (GS) transformation and Principle Analysis component transformation (PCA), Color normalized (CN) transformation and PAN Sharpening transformation 6 kinds of fusion methods fuse multi-spectral images and panchromatic images. Finally with image gray value of the mean, standard deviation, entropy, average gradient, correlation coefficient. These 5 indexes evaluate the fused images objectively. As a result, the best fusion method in the study areas is Gramm-Schmidt (GS) transformation, which preserve the spectral information and enhance the spatial information of images in the study area.(2)Compared with the traditional SVM classification method, investigation on Jinling village, TianMa townin dujiangyan city of plain region in ChengDu has found out that the accuracy of these 4 filed boundary extraction methods is higher from 1.72% to 2.79% and the Kappa coefficient is higher from 0.034 to 0.052, which shows pretty good boundary alignment and continuity. The accuracy of panchromatic Sobel detection combined with Object-oriented method classification (95.12%) and Kappa coefficient (0.938) is highest, followed by the Object-oriented classification technology, and Roberts detection combined with Object-oriented method has more "wrongly classification" and "leakage classification" phenomenon, the HP detection combined with Object-oriented method of the "wrongly classification" and "leakage classification" is more serious. Therefore, panchromatic Sobel detection combined with Object-oriented method is the optimal method of field boundary extraction in this region.(3)Compared with the traditional SVM classification method, investigation on hilly area in basin of Longxuvillage, Qi long town, FuShun country has found out that the accuracy of these 4 filed boundary extraction methods is higher from 2.7% to 5.53% and the Kappa coefficient is higher from 0.03 to 0.065, which shows good boundary alignment and continuity. The accuracy of panchromatic Sobel detection combined with Object-oriented method classification (94.65%) and Kappa coefficient (0.923) is the highest, followed by the Object-oriented classification technology, and Roberts detection combined with Object-oriented method has more inner cavity, which cause serious boundary fracture, the HP detection combined with Object-oriented method of the "wrongly classification" and "leakage classification" is more serious. Therefore, panchromatic Sobel detection combined with Object-oriented method is the optimal method of field boundary extraction in hilly area in basin.(4) The extraction results on 4 kinds of extraction methods of mountain areas around the SiChuan basin of Guang Hui village, GuLin town,GuLin county is the worst, which shows low field boundary of inosculation, poor continuity, mainly influenced byphase of image, ground objects and the environment.Compared with the traditional SVM methods, only the object-oriented classification technology of classification accuracy and Kappa coefficient is 5.04% and 0.051 higher respectively, and the remaining three combination methods are low. In which the classification accuracy of object oriented technology (80.19%) and Kappa coefficient (0.703) is the highest, but some "wrongly classification" and "leakage classification" phenomenon till exist. The similarity of the extracted boundary and the actual boundary is low; the rest three methods have more serious "wrongly classification" and "leakage classification" phenomenon. Therefore, all kinds of extraction methods were not suitable for field boundary of extracting in mountain area around the SiChuan basin. |
PDF Full Text Request