Distribution Extraction of Mikania micrantha Based on UAV Hyperspectral Image: A Case Study in Dehong, Yunnan Province, China

Abstract:

As a highly dangerous alien species, Mikania micrantha has become a serious threat to the ecosystem health and biodiversity of invasive sites. In order to effectively control its invasion, and grasp its spatial distribution and dynamic change, its distribution in Dehong Prefecture, Yunnan Province was extracted by deep learning (DL), support vector machine (SVM) and random forest (RF) methods based on UAV hyperspectral data. The results showed that three methods could effectively extract the distribution of M. micrantha, in which DL method had the best extraction effect with mapping accuracy and user accuracy of 96.61% and 95.00%, respectively, followed by the RF method with those of 94.83% and 91.67%, and the SVM method with those of 92.45% and 81.67%. All three methods could well extract the concentrated distribution areas of M. micrantha, the methods of DL and RF were better than SVM in identification of fragmented distribution of M. micrantha. Therefore, UAV hyperspectral images would provide supports and basis for the monitoring, early warning and precise control of M. micrantha invasion, which was of great significance to protect the security of local ecosystems.

Key words:Mikania micrantha;UAV remote sensing;Deep learning;Support vector machine;Random forest