Method Of Extracting Canal Distribution Based On UAV Remote Sensing System

Irrigation canal system is the main water conservancy facilities in large-scale irrigation district. The realization of extracting irrigation canal distribution will have an important significance on agricultural water transmission efficiency, dynamic monitoring of irrigation information and the improvement of irrigation decision management. In this paper, the UAV was used as a platform equipped with visible light camera and multi-spectral camera to obtain the high resolution remote sensing images of irrigation canal system. After analyzing the characteristics of the irrigation canal system, extracting the canal distribution respectively from the study area on the optical images and multispectral remote sensing images. By the means of experimental verification, completeness and preciseness were used to evaluating the extraction results. The object oriented method to extract the irrigation canal distribution from multi spectral remote sensing images was more complete and accurate. The main research contents and conclusions as follows:1. Important irrigation canal distribution of the study area had been investigated. The remote sensing images were obtained by aerial photography with fixed-wing UAV equipped visible light camera and quad-rotor UAV mounted multi-spectral camera. After the remote sensing image preprocessing, irrigation canal characteristics obtained by analyzing and summarizing. As is the trench structure color features were displayed as dark, dim brightness in the image, geometric characteristics showed little change within a certain range in curvature of the linear characteristic, spectral characteristics were basically the same after Band Math, topological characteristics were continuity and not interrupted suddenly.2. The K-means clustering was used to realize the image segmentation in the visible light image. Using the irrigation canal color characteristics setted the initial number of cluster centers 3 through iteration clustered the image into three categoriesthe to obtain the segmentation image. Mathematical morphology method was used in the postprocessing to extract the irrigation canal system, connect the broken canal system and refine the canal system. By the means of experimental verification and accuracy evaluation, the completeness and preciseness of the results in visible light image extraction were obtained as 0.55 and 0.5respectively. From the two points of the interference factor and extraction method, it wasconcluded that the extraction had a very good result on the small area with simple canal system for the visible light image. But it was not ideal for large scale irrigation district with complex features and canal system distribution.3. Object oriented method was selected for multi spectral remote sensing image. Some rules were established through the spectral and geometric features of irrigation canal system to achieve the extraction of irrigation canals, other rules were established through the spectral of water and geometric features of irrigation canal system to achieve the extraction of irrigation canals ongoing operation. Through the combination of each extraction realized the extraction of irrigation canal distribution and obtained the vector data. The vector results of irrigation canal system were obtained through the modification of the vector data. While the completeness and preciseness of the results in multi spectral image extraction were 0.75 and0.95. Experimental results showed that using the spectral and geometry characteristics of the irrigation canal system, oriented object method could effectively solve the phenomenon "Synonyms object with different spectrum, and different objects with synonyms spectrum" in the multi spectral image and the obtained results were satisfactory.4. The effective combination of different features in remote sensing image of main canals,branch canals, agricultural drainage, watered canals and anhydrous canals in the study area could achieve the extraction of complete irrigation canal system. Experimental results showed that the proposed method was effective and feasible and could extract high quality irrigation canal distribution information from the multispectral remote sensing images.