Comparative Study on Classification Methods of Typical Wetland Features in UAV Images
Aiming at the centimeter-level resolution UAV images, a wetland park in Wuxi is selected as the research object. Firstly, the wetland image is segmented at multiple scales, and the ESP tool is used to obtain the best segmentation parameters. Then, the fea-ture is selected, and three classification methods such as decision tree (DT), Bayesian (Bayes) and random forest (RF) are selected to classify the typical wetland features. The classification results and accuracy of different methods are compared and analyzed. The comparison results show that the accuracy of random forest algorithm is the highest in the classification of typical wetland elements, and the accuracy of decision tree and Bayesian classification algorithm is inferior to that of random forest. In terms of classification effi-ciency, the random forest algorithm takes the longest time and involves parameter setting adjustment, while the efficiency of Bayesian algorithm is much higher than that of decision tree and random forest. Moreover, the algorithm has simple operation, no parameter set-ting and is easy to be applied in production.
万方数据
维普
