Benthic and substrate cover data derived from field photo-transect surveys for the Townsville to Whitsunday management region of the Great Barrier Reef (GBR), May/June 2019

A subset of photoquadrats were uploaded to the CoralNet machine learning interface (https://coralnet.ucsd.edu/) and manually labelled for coral, algae or substrate type using 50 points per quadrat. Follow training of the machine, this training set enabled automatic annotation of all unclassified field images: the remaining field photos were uploaded to the database and 50 annotation points were overlaid on each of the images. Every point was assigned a benthic cover category from a label list automatically by the program. The resulting benthic cover data of each photo was linked to GPS coordinates, saved as an ArcMap point shapefile, and projected to Universal Transverse Mercator WGS-84.

本数据集包含部分样方照片（photoquadrats），已上传至CoralNet机器学习交互界面（https://coralnet.ucsd.edu/），并按照每个样方50个标注点的规则，针对珊瑚、藻类及基质类型完成手动标注。完成该机器学习模型的训练后，此训练集可实现对所有未分类野外图像的自动标注：剩余野外照片上传至数据库后，将在每张图像上叠加50个标注点，程序将自动从预设标签列表中为每个标注点分配底栖覆盖（benthic cover）类别。最终，每张照片对应的底栖覆盖数据将与GPS坐标关联，保存为ArcMap点形状文件，并投影至通用横轴墨卡托WGS-84坐标系（Universal Transverse Mercator WGS-84）。