Research On Extraction Method Of Terraced Field Based On Improved AdaBoost

Terrace is an important measure to control soil erosion in Loess Plateau.At present,there has been a certain research progress in terrace extraction from high-resolution terrace images obtained by UAV,but in the previous research,the factors of many features and complex terrain in terrace area are not considered,so the extraction effect is poor.The current extraction algorithm can not adapt to different terrain features of terrace area.In order to solve the problem of poor adaptability of the current extraction algorithm,this thesis studies the extraction of terrace area based on the improved AdaBoost method.This thesis mainly completes the work as follows:(1)Data set generation and preprocessing.Firstly,the image defogging algorithm is used to enhance and restore the high-resolution Orthophoto Image of UAV and improve the image clarity;Then,six common terrain parameters,such as slope,roughness and height coefficient of variation,are calculated by digital elevation model,and the slope,roughness,mountain shadow and slope of slope are obtained as terrain parameters with low correlation;The Orthophoto Image and terrain parameters are fused to get the image rich in terrain information.The multi-scale segmentation algorithm is used to segment the image into multiple objects,and the objects are transformed into sample data to generate the sample data set;Finally,the sample equalization is used to expand a small number of samples to achieve the equalization,and the preprocessing of the sample data set is completed.(2)Based on the comparison of traditional AdaBoost and common terrace region extraction methods.In this thesis,the traditional AdaBoost algorithm is used to extract the terrace region,and the weight of the base classifier is recalculated by the change of sample weight in the iterative process.Finally,10 single decision trees are integrated to realize the construction of the model.A total of 1304 samples of dense strip area,irregular area and sparse block area are selected to test and train the model according to the ratio of 6:4,and compared with the KNN and Naive Bayes and logistic Seven common terrace extraction algorithm are compared.The results show that the traditional AdaBoost algorithm has better extraction effect than other seven common terrace extraction algorithms.(3)Research on terrace region extraction method based on improved AdaBoost.In order to improve the extraction effect of traditional AdaBoost algorithm,the weight distribution strategy of AdaBoost is improved.Four new weight distribution strategies are introduced,and the extraction results of the four weight distribution strategies are compared and analyzed.When the weight distribution strategy is improved to n~2,the extraction effect of the three research areas is better.The results show that the improved AdaBoost algorithm can improve the kappa coefficient and total accuracy by 0.0835,3.34%and 0.7890,90.91%of dense strip area,irregular area and sparse block area respectively.The research of this thesis promotes the research and application of ensemble learning in remote sensing image terrace region extraction,and promotes the development of smart agriculture.