Universidade Federal Rural da Amazonia (UFRA), CP. 917, Belém, Pará, 66077-530, Brazil; Universidade Federal do Sul e Sudeste do Pará (UNIFESSPA), Folha 31, Quadra 07, Lote Especial, Nova Marabá, 68507-590, Marabá, Brazil.
Institut de Recherche pour le Développement (IRD), UMR 228 ESPACE DEV, 500, Rue Jean François Breton, 34093, Montpellier, France.
J Environ Manage. 2017 May 15;193:40-51. doi: 10.1016/j.jenvman.2017.02.004. Epub 2017 Feb 9.
High spatial resolution images as well as image processing and object detection algorithms are recent technologies that aid the study of biodiversity and commercial plantations of forest species. This paper seeks to contribute knowledge regarding the use of these technologies by studying randomly dispersed native palm tree. Here, we analyze the automatic detection of large circular crown (LCC) palm tree using a high spatial resolution panchromatic GeoEye image (0.50 m) taken on the area of a community of small agricultural farms in the Brazilian Amazon. We also propose auxiliary methods to estimate the density of the LCC palm tree Attalea speciosa (babassu) based on the detection results. We used the "Compt-palm" algorithm based on the detection of palm tree shadows in open areas via mathematical morphology techniques and the spatial information was validated using field methods (i.e. structural census and georeferencing). The algorithm recognized individuals in life stages 5 and 6, and the extraction percentage, branching factor and quality percentage factors were used to evaluate its performance. A principal components analysis showed that the structure of the studied species differs from other species. Approximately 96% of the babassu individuals in stage 6 were detected. These individuals had significantly smaller stipes than the undetected ones. In turn, 60% of the stage 5 babassu individuals were detected, showing significantly a different total height and a different number of leaves from the undetected ones. Our calculations regarding resource availability indicate that 6870 ha contained 25,015 adult babassu palm tree, with an annual potential productivity of 27.4 t of almond oil. The detection of LCC palm tree and the implementation of auxiliary field methods to estimate babassu density is an important first step to monitor this industry resource that is extremely important to the Brazilian economy and thousands of families over a large scale.
高空间分辨率图像以及图像处理和目标检测算法是辅助生物多样性研究和商用林木种植的新技术。本文旨在通过研究随机分散的原生棕榈树,为这些技术的应用提供相关知识。文中,我们利用巴西亚马孙地区一个小型农业社区的高空间分辨率全色 GeoEye 图像(0.50 米),分析了大型圆形树冠(LCC)棕榈树的自动检测。此外,我们还提出了基于检测结果估算 Attalea speciosa(巴巴苏)棕榈树 LCC 密度的辅助方法。我们利用基于数学形态学技术检测开阔区域棕榈树阴影的“Compt-palm”算法,并结合实地方法(即结构普查和地理参考)验证空间信息。该算法能够识别 5 级和 6 级的个体,通过提取百分比、分枝因子和质量百分比因子来评估其性能。主成分分析表明,所研究物种的结构与其他物种不同。大约 96%的 6 级巴巴苏个体被检测到,其叶柄明显小于未被检测到的个体。而 5 级巴巴苏个体中,有 60%被检测到,其总高度和叶片数量与未被检测到的个体明显不同。我们对资源可用性的计算表明,6870 公顷土地上有 25015 株成年巴巴苏棕榈树,杏仁油年潜在生产力为 27.4 吨。LCC 棕榈树的检测以及辅助实地方法估算巴巴苏密度的实施,是监测这一对巴西经济和数千个家庭具有重要意义的产业资源的重要的第一步。