Tree species detection with UAV-based hyperspectral imagery

当收获后重新种植树木时，必须对林地进行早期**再生期间的监测和管理，以确保用与收获的树种相似的物种组合重新建立森林，之后**这些地块被认为可以自由生长到成熟。First Resources Management Group Inc.(FRMG)为森林保有者提供**可持续森林管理服务，这些服务包括将重新种植的森林**带到这个自由生长的状态。为此，FRMG开发了遥感技术，利用卫星图像、从固定翼平台获得的照片和激光雷达数据**以及从直升机上拍摄的照片来监测**林地。这使得FRMG能够以良好的**精度绘制a)树干密度和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，因为近年来加拿大发生了几起与林业有关的直升机坠毁事件，导致**飞行员和机组人员死亡。其次，用现有的混合数据来源**确定树种不够准确，因此需要昂贵的实地数据收集作为补充。本项目旨在测试两个新的数据来源，并将其整合到现有技术组合中：a)基于无人机的**高光谱图像，以及b)非常高分辨率的卫星图像，并开发用于物种识别的自动化数据处理**算法。我们将测试这些新数据源如何单独识别树种**，与现有数据源相结合，以及与其他**数据源相结合，但不包括直升机图像。该项目将使用安大略省两个现有**林地的实地观测数据处理算法的校准和验证，并将使林业局能够**优化其基于遥感的林地监测方法。这一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**这项研究的基础是安大略省，但它是在加拿大乃至全球范围内应用的。

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.