| Cultivated land is the valuable resource for mankind to live by.It is an important foundation for protecting the people's living standards and country's stability to grasp the information of farmland timely and dynamically.How to extract the cultivated farmland information accurately has become an important research field for monitoring cultivated land.The Unmanned Aerial Vehicle(UAV)remote sensing platform offers many advantages.It has been used on a large-scale in the southwest of China for the past few years.Although the UAV remote sensing image provides lots of information,but how to extract the farmland information is an important and difficult study content.In the paper,the image is segmented by multi-scale segmentation method with Canny edge detection information based on different bands.Then the image is classified by Nearest Neighbor Classification and the classification based on SEaTH algorithm.This thesis research and explore the methods of extracting farmland information.Main productions and results are as follows:(1)The optimal split scale of the research area:The scale of 130 is more appropriate for the segmentation of vegetation and non,vegetation.The scale of 110 is more appropriate for the segmentation of vegetation and vegetation farmland.The scale of 85 is more appropriate for the segmentation of non-vegetation,building,concrete pavement,water and film-mulch farmland.(2)It can achieve good results,when the algorithm based on the edge detection of Canny be used for aided segmentation image.It will get different results for Canny edge operators based on different bands.Canny G(based of the band of Green)is more appropriate for segmenting woodland and vegetable farmland.Canny R(based on the band of(Red)is more appropriate for segmenting non-vegetation,building,concrete pavement,water and film-mulch farmland.(3)Different homogeneity factors also affect the final segmentation result.In this paper,the suitable weight of spectrum are 0.1 and 0.2,the suitable weight of compactness and smoothness are both 0.5.(4)The omission errors are comparatively high(35.3%,36.1%)when the land cover of vegetable farmland and non-vegetation by the method of are classified with Nearest Neighbor Classification.The omission error and misclassification error are both comparatively high the land cover of film-mulch farmland are classified by the method of classification based on SEaTH algorithm(33.7%?42.9%).The omission errors of farmland and non-vegetable are obvious decreased through the method of Nearest Neighbor Classification combined with the classification method based on SEaTH algorism.The user accuracy of film-mulch farmland is higher 36.3%and 25.6%than that the other two methods.(5)The method of Nearest Neighbor Classification combined with the classification method based on SEaTH algorism overall accuracy(89.3%)is higher than that of the classification method based on SEaTH algorism(80.7%),is higher than the method of Nearest Neighbor Classification(78.7%).The method of Nearest Neighbor Classification combined with the classification method based on SEaTH algorism kappa coefficient(0.83)is higher than that of the classification method based on SEaTH algorism(0.69),is higher than the method of Nearest Neighbor Classification(0.67).The classification method based on SEaTH algorithm is more appropriate to classify the two kinds of land covers with a big difference features.The classification method of Nearest Neighbor Classification is better at classifying complex land covers.(6)It's appropriate to classify vegetation land and non-vegetable land with the method of Nearest Neighbor Classification.Extract the land cover of vegetable farmland from vegetation land with SEaTH algorism.Extract the land cover of non-vegetable farmland from non-vegetation land with the method of Nearest Neighbor Classification.Extract the land of concrete pavement and film-mulch farmland with Nearest Neighbor Classification.And then,extract film-mulch farmland with SEaTH algorism. |
PDF Full Text Request