Research On The Classification Method Of Land Use Type Based On UAV Remote Sensing In Irrigation District

In order to verify the availability of UAV(Unmanned Aerial Vehicle) optical remote sensing technology in irrigation district land use types and classification, WuYuan County taerhu town area was chose as the research area. The visible images were obtained by using TEZ fixed wing UAV equipment with SONY A5100. To build different classification methods were extracted the land use types by using object-oriented method. Confusion matrix were evaluated the accuracy of classification results. The best method based on the UAV remote sensing system to obtain the high resolution visible light image data was determined. The main content and conclusions of the paper were as follows.(1)The 150 images data of original visible high-resolution by using the UAV remote sensing system, were mosaicked in the Agisoft PhotoScan software. Output resolution of0.1m orthophoto data, it could reflect the features of the ground objects in the test area. With the specific situation of study area, the classification system was constructed, and established thetest area land use types.(2)The digital orthophoto data was adopted object oriented, and utilized multi-scale segmentation technology. By means of trial and error method, except water and water conservancy facilities land and special land, the UAV remote sensing image data of optimum parameters were determined that segmentation scale parameter was 300; the shape factor parameter was 0.4; and the close degree parameter was 0.5.It was capable of remaining intact throughout the division.(3)Decision Tree, different kernel function of Support Vector Machine, and K-Nearest Neighbor Classification were established to extract land use types according to the specificity of ground object in spectrum characteristics, shape characteristics and texture characteristics.Confusion matrix were applied to evaluated the accuracy of classification results. Results indicate that RBF kernel function of Support Vector Machine can accurately extract the characteristics of ground object, the overall accuracy is 82.20%, kappa coefficient is 0.7659;overall accuracy and kappa coefficient of linear kernel function Support Vector Machine are respectively 81.40% and 0.7564; overall accuracy and kappa coefficient of Decision Tree respectively are 74.00% and 0.6675; overall accuracy and kappa coefficient of K-NearestNeighbor Classification respectively are 74.00% and 0.6675. On the basis of water and water conservancy facilities and special land use were extracted byvisual.(4)Based on the RBF kernel function of Support Vector Machine classification method combined with the Decision Tree model, the overall accuracy is 84.20%, Kappa coefficient is0.7900. It indicated that the way of combining the two classification methods was feasible for the land use types classification for remote sensing images by UAV. And support vector machine classification of water and water conservancy facilities land for automatic extraction,Its producer’s precision is 66.67%, and the user’s precision is only 43.48%. The results show that the visible light remote sensing technology can be used to extract the land use type of the irrigation area, but the extraction of the ditch needs further study.