Automated classification of douglas-fir beetle infested trees in Unpiloted Aerial Vehicle (UAV) acquired imagery**

在不列颠哥伦比亚省南部，及早发现道格拉斯冷杉甲虫(IBD)侵扰越来越重要，那里的IBD出没面积已从2013年的约7000公顷增加到2017年的近8万公顷。IBD感染，通常会迅速杀死树木，当木材被收获或未被收获时，木材的价值降低，从而导致可用于维持森林火灾的燃料。目前，IBD感染的检测是通过直升机俯瞰飞行期间目测森林树冠和现场工作人员从地面目测个别树木来进行的。直升机俯瞰可以覆盖大片区域，但不能量化IBD感染的规模，因为人眼无法在感染的早期阶段(绿色攻击)感知树冠的变化，也因为人眼无法从直升机俯瞰飞行的高度分辨出个别树木。因此，直升机俯瞰不能产生死亡(灰色攻击)或垂死(红色攻击)树木的准确计数。另一方面，实地走访可以准确地识别绿色、红色和灰色发病阶段的IBD感染，但现场方法在经济上不能很好地推广到大范围。因此，目前用于量化IBD感染的方法是不够的，导致当IBD感染的树木仍然具有经济价值时，它们的收获不完整。通过与汤普森·里弗斯大学的研究人员合作，Second Pass林业有限公司(SPF)的林业专业人员希望开发和验证一种半自动系统，用于分析无人驾驶飞行器(UAV)获取的多光谱图像，以识别IBD攻击的所有三个阶段(绿色、红色和灰色)的树木。从这项研究中获得的知识将与实施图像分析工作流程的软件一起转移到SPF。通过在同行评议的期刊上发表这项研究的结果，从这项研究中获得的知识将更广泛地传播给林业社区。

Early detection of Douglas fir beetle (IBD) infestation is increasingly important in southern British Columbia, where the IBD infested area has grown from approximately 7,000 ha in 2013 to almost 80,000 ha in 2017. IBD infestation, generally kills the tree rapidly reducing the value of the timber when it is harvested or if is not harvested, contributing to the fuel available to sustain forest fires. Currently, detection of IBD infestation is carried out by visual inspection of the forest canopy during helicopter overview flights and by visual inspection of individual trees from ground level by field crews. Helicopter overviews can cover large areas but cannot quantify the scale of IBD infestation, because the human eye cannot perceive changes in the canopy during the early stage (green attack) of the infestation, and because the human eye cannot resolve individual trees from elevation of a helicopter overview flight. Thus, helicopter overviews cannot produce an accurate count of dead (grey attack) or dying (red attack) trees. Field visits, on the other hand, can accurately identify IBD infestation in the green, red, and grey attack stages, but field method do not scale well economically to large areas. Thus, current methods for quantifying IBD infestation are insufficient, resulting in an incomplete harvest of IBD infested trees when they are still economically valuable. Through a partnership with researchers at Thompson Rivers University, forestry professionals at Second Pass Forestry Ltd. (SPF) want to develop and validate a semi-automated system for analyzing unpiloted aerial vehicle (UAV) -acquired multispectral imagery to identify trees in all three stages (green, red, and grey) of IBD attack. The knowledge gained from this research will be transferred to SPF along with software implementing the image analysis workflow. Through publication of the outcomes of this research in a peer-reviewed journal, the knowledge gained from this research will be distributed more broadly to the forestry community.******