Tree species detection with UAV-based hyperspectral imagery

当树木在收获后重新种植时，必须在早期再生期间对森林地块进行监测和管理，以确保重新建立的森林具有与收获的物种相似的物种组合，之后地块被认为可以自由生长直到成熟。First Resource Management Group Inc. （FRMG）为森林权属所有者提供可持续森林管理服务，包括将重新种植的森林地块带到这个自由生长的州。为此，FRMG开发了基于遥感的技术，利用卫星图像、从固定翼平台获取的照片和激光雷达数据，以及从直升机上拍摄的照片，对森林地块进行监测。这使得FRMG能够以良好的精度绘制a)茎密度和b)树高，并以合理的精度绘制c)树种。然而，目前的方法有两个主要缺点。最重要的是，从直升机上获取低空图像有相当大的生命风险，因为近年来在加拿大发生的几起与森林有关的直升机坠毁事故导致飞行员和机组人员死亡。其次，利用现有的混合数据来源对树种的识别不够准确，因此需要昂贵的实地数据收集作为补充。目前的项目旨在测试并将两种新的数据源集成到现有的技术组合中：a)基于无人机的高光谱图像，b)非常高分辨率的卫星图像，并开发用于物种识别的自动数据处理算法。我们将测试这些新数据源如何单独识别树种，如何与现有数据源结合，如何与其他数据源结合，但不包括基于直升机的图像。该项目将使用来自安大略省两个现有森林样地的实地观测来校准和验证数据处理算法，并将使FRMG能够优化其基于遥感的森林样地监测方法。研究结果还将使森林管理部门更广泛地了解遥感技术在森林监测方面的潜力。虽然这项研究是在安大略省进行的，但它在加拿大乃至全球范围内都得到了应用。

When trees are replanted after harvest, forest plots must be monitored and managed for the period of early**regrowth to ensure the forest is re-established with a species mix similar to that harvested, after which the plots**are considered free to grow until maturity. First Resource Management Group Inc. (FRMG) provides**sustainable forest management services to forest tenure holders, services that include bringing re-planted forest**plots to this free-to-grow state. To this end, FRMG has developed remote sensing-based technology to monitor**forest plots using a combination of satellite imagery, photos and lidar data acquired from fixed-wing platforms,**and photos taken from helicopter. This allows FRMG to map a) stem density and b) tree height with good**accuracy, and map c) tree species with reasonable accuracy. However, the present approach suffers from two**principal drawbacks. Most importantly, there is considerable risk to human life involved with the acquisition of**low-altitude imagery from helicopter, as several forestry-related helicopter crashes in Canada have resulted in**the deaths of pilots and crew in recent years. Secondly, the identification of tree species with the existing mix**of data sources is not sufficiently accurate, and it is therefore complemented by costly field data collection. The**present project aims to test and integrate two new data sources into the existing technology mix: a) UAV-based**hyperspectral imagery, and b) very high resolution satellite imagery, and to develop automated data processing**algorithms for species identification. We will test how these new data sources allow identification of tree**species on their own, in combination with the existing data sources, and in combination with the other data**sources but excluding the helicopter-based imagery. The project will use field observations from two existing**forest plots in Ontario for calibration and validation of data processing algorithms, and will allow FRMG to**optimize their remote sensing-based approach to forest plot monitoring. The result will also inform the forest**management sector more broadly about the potential of remote sensing technology in forest monitoring. While**the research is based in Ontario, it is application across Canada, and globally.