1:45 PM - 3:15 PM
[HGG01-P10] Exploring supervised classification of land cover using airborne lidar and aerial imagery
Keywords:land cover classification, airborne lidar, aerial imagery, decision tree, support vector machine
The purpose of this study is to use airborne lidar and aerial imagery data for land cover classification. By extracting the geometric characteristic parameters of the airborne lidar data and the spectral characteristic parameters of the aerial images, the object-oriented classification software is used to analyze and classify the characteristics of the ground objects. The ground objects in the research area are divided into six categories: artificial ground, buildings, bare land, grassland, trees and planted shrubs. The research is divided into three stages. The first stage is to use the elevation difference, dispersion, and wave width extracted from the lidar data to classify the land cover, and then divide the parameters into the elevation difference to perform the decision tree classification and accuracy evaluation. In the second stage, in addition to using LiDAR data, aerial images are added for decision tree classification and accuracy evaluation, hoping to improve the accuracy of classification. The third stage is to use the support vector machine classification method for classification, so as to compare the classification results of different classification methods.
In this study, aerial image data and lidar data were used, supplemented by decision tree and support vector machine algorithms for land cover classification, with the highest overall accuracy of 84%. It can be seen that the relevant parameters and classification methods selected in this study have achieved good classification results and can be effectively used in the task of ground object recognition. Comparing the classification results of each experimental group, the producer accuracy and user accuracy of the three categories of trees, buildings, and planted shrubs are in the range of 43-97%. The manufacturer's accuracy and user's accuracy of artificial ground, grassland and bare land are in the range of 20-72%. The reason for this analysis is that the former three types have large differences in characteristics such as spectrum, elevation difference, dispersion, and wave width, while the characteristics of the latter three types are less obvious. Therefore, misjudgment of classification is easy to occur, resulting in poor classification results.
In this study, aerial image data and lidar data were used, supplemented by decision tree and support vector machine algorithms for land cover classification, with the highest overall accuracy of 84%. It can be seen that the relevant parameters and classification methods selected in this study have achieved good classification results and can be effectively used in the task of ground object recognition. Comparing the classification results of each experimental group, the producer accuracy and user accuracy of the three categories of trees, buildings, and planted shrubs are in the range of 43-97%. The manufacturer's accuracy and user's accuracy of artificial ground, grassland and bare land are in the range of 20-72%. The reason for this analysis is that the former three types have large differences in characteristics such as spectrum, elevation difference, dispersion, and wave width, while the characteristics of the latter three types are less obvious. Therefore, misjudgment of classification is easy to occur, resulting in poor classification results.