Surface Extraction Of Terraced Field From UAV Imagery Based On Object-oriented Analysis

As a major measure for the improvement of slope farmland,the construction of terraced fields plays a key role in the development of agriculture and the control of soil erosion in loess hilly regions in China.Accurate and efficient extraction of terraced fields can provide reliable technical services for improving agricultural economic development in the Loess Hilly Region and assessing terrace construction.At present,semi-automated interpretation of terraced fields information through remote sensing imageries had some progress,but it was limited by the cost,precision,single data source and single interpretation method of data acquisition.The study on terrace information extraction was confined to large range.Accurate extraction of terraces at low cost still require further study.The experimental area was a horizontal dry terraced field in Longquan Township,Yuzhong County,Lanzhou,Gansu Province.Based on ObjectOriented Image Analysis(OBIA),using UAV remote sensing imagery and terrain factor calculated by DEM,study on the approach of extracting terraced field surface.The main work of the paper is as follows:(1)Study on terrace segmentation.Firstly,imagery enhancement was performed on UAV orthoimage(resolution 0.5m).The terrain factors were calculated by a Digital Elevation Model(DEM)(resolution 0.5m)and dimension reduction.Secondly,the enhanced imagery,terrain factors dimension reduced,and the fusion of the two were segmented for obtaining the terrace objects and non-terrace objects using multi-scale segmentation and optimal scale selection theories.The results show that using the data that combined of orthoimage and terrain factor is better than using a single data source.(2)Study on classification of terraced field surface object.The surface of terraced field was extracted by supervised learning(K-nearest,support vector machine and decision tree)for the terraced objects and non-terrace objects obtained by multiscale segmentation.The accuracy of the classification was evaluated using the total precision and Kappa coefficient of the calculation based on the confusion matrix method.It is showed that the classification results based on the fusion of orthoimages and terrain factors is better than that based on a single data source.Furthermore,for the terraced fields with more regular shape and less surface coverage,the classification results based on support vector machines are superior to the K-nearest neighbor method and the decision tree method.For the terraced fields with irregular shape and snow cover in the field,the classification results based on the decision tree method are superior to the Knearest neighbor method and the support vector machine method.