Benthic hard coral point score estimates for Barracouta shoal, North West Shelf from 2010, 2011, 2013 and 2016

These data consist of 20-point score estimates randomly placed on individual high resolution downward facing benthic digital images taken at Barracouta shoal using the AIMS towed video system.The AIMS towed video system comprises a towed camera platform sending a live camera feed to a vessel-based, realtime image classification system (see Heyward et al. 2011) and a downward-facing high resolution still camera and strobe system programmed to take sequential still images at fixed time intervals of 10 seconds. The towed platform was deployed over the stern of the vessel, maintained as near as possible within a metre of the seabed and towed at 1-2 knots (1.5 nominal). Transect lengths varied among the years of data collection.The downward-looking still images were geo-referenced during post-processing then analysed using a point-intercept approach. Information on benthic biota at each shoal was extracted from images using a point intercept approach with the AIMS Reefmon software (Jonker et al., 2008). All images were analysed using the Reefmon database system, with five overlaid points classified per photo and data logged against transect, depth and position.The data provided here are derived using a machine learning model trained using the original manual annotations. The artificial intelligence engine called BenthoBot was used to re-analyse all seabed images from all years 2010-2016, processing each image using exactly the same approach. BenthoBot is a computer algorithm developed to classify points on an image, based on the spectral properties extracted from each image. It has been developed specifically by the Australian Institute of Marine Science to providean efficient and consistent means of generating the point based broad scale benthic classification data. The benefits of using BenthoBot include standardisation of the number of points sampled per image across all years (20 points per image) and removal of inconsistency in point classification associated with numerous technicians scoring images that may cause spatial and temporal artifacts.Secondary (textural) datasets correlated with seafloor properties were developed from multibeam bathymetry to provide information on environmental characteristics, and are also provided here extracted for each image location as covariates.

本数据集包含20个点位评分估算值，这些点位随机布设于巴库塔沙洲（Barracouta Shoal）采集的高清俯拍底栖数字图像之上，数据采集采用澳大利亚海洋科学研究所（Australian Institute of Marine Science, AIMS）拖曳视频系统。
该AIMS拖曳视频系统由拖曳式摄像平台构成，该平台可将实时摄像画面传输至船载实时图像分类系统（详见Heyward等人2011年研究），同时搭载一台俯拍高清静态相机与频闪系统，该系统按固定10秒间隔连续拍摄静态图像。
拖曳平台从船尾布设，尽可能保持距海底1米以内的高度，以1-2节（额定1.5节）的航速拖曳。数据采集各年份的样带长度存在差异。
俯拍静态图像在后期处理阶段完成地理配准，随后采用点截距法开展分析。研究人员借助AIMS Reefmon软件，通过点截距法从图像中提取各沙洲的底栖生物信息（Jonker et al., 2008）。
所有图像均通过Reefmon数据库系统完成分析，每张照片叠加5个点位进行分类，并将数据关联记录至样带、水深与位置信息中。
本次发布的数据集源自经原始人工标注训练得到的机器学习模型。研究人员采用名为BenthoBot的人工智能引擎，对2010-2016年所有年份的海底图像进行重新分析，且对每张图像采用完全一致的处理流程。
BenthoBot是一种基于单幅图像提取的光谱特性对图像点位进行分类的计算机算法。该算法由澳大利亚海洋科学研究所专门开发，旨在提供一种高效且一致的方法，生成基于点位的大规模底栖分类数据。
使用BenthoBot的优势包括：统一各年份每张图像的采样点位数量（每张图像20个点位），同时消除因多名技术人员对图像进行评分而导致的点位分类不一致问题——这类不一致可能引发空间与时间层面的人为误差。
研究人员从多波束测深数据中提取与海底特性相关的次级（纹理）数据集，以提供环境特征信息，本次发布的内容中也包含了各图像点位对应的此类协变量数据。