干旱区生态环境知识资源中心

Remote sensing from unmanned aerial vehicles provides an opportunity to bridge the gap between fine scale ground-based measurements and broad scale observations from conventional aircraft and satellites. The advantages of this approach include safe access to hazardous or difficult terrain and conditions, the ability to survey at specific times and spatial scales, and the increasingly affordable cost of this technology. These benefits have led to a rapidly expanding range of applications in natural resource management and research including mapping of terrain, vegetation cover and condition, threatened species, habitat and the impacts of agriculture, forestry, urbanisation and climate change. The analysis of these often large datasets requires reliable segmentation and classification algorithms to efficiently process information for use in landscape ecology and adaptive management. In this paper, four segmentation methods are compared using images of native vegetation, introduced weeds and agriculture recorded from a quadcopter flown over a warm temperate island (Waiheke Island, New Zealand), and also images recorded from a fixed wing UAV in a polar desert (McMudro Dry Valleys, Antarctica). We propose a post-processing method to improve the segmentation performance of the algorithms and demonstrate how this can contribute to improving research outcomes in natural resource management, conservation and agriculture.