Risso’s dolphin dataset

Photo identification (photoID) is a non-invasive technique devoted to the identification of individual animals using photos, and it is based on the hypothesis that each specimen has unique features useful for its recognition. This technique is particularly suitable to study highly mobile and hard to detect marine species, such as cetaceans. These animals play a key role in marine biodiversity conservation because they maintain the stability and health of marine ecosystems due to their apical role as top predators in food webs. The information obtained with photo identification studies is useful for acquiring new knowledge on their abundance estimation, social dynamics, pattern migration and site fidelity of the target species. Since manual photo identification is time-consuming and impractical in cases of large datasets, the employment of advanced automated techniques can support users and accelerate the process of individual photo identification. The availability of open datasets, such as the dataset presented here and entitled “Risso’s dolphin dataset”, can contribute to the development and implementation of fully automated cetacean photo identification techniques, either by acting as training datasets, or by providing already identified individuals for possible matching. In general, it can be useful for the development of a new methodology based on image processing, computer vision and machine learning, devoted to image cropping, segmentation, and recognition. Regarding the species object of this dataset, the Risso’s dolphin Grampus griseus (Cuvier, 1812), is one of the least-known cetacean species on a global scale, with Mediterranean subpopulation ranked as Endangered by the IUCN Red List. Risso’s dolphins exhibit long-lasting identifiable natural marks on their dorsal fin, and these patterns make the species particularly suitable for photoID algorithms. Hence, to bridge the gap to understanding this species, a key component is obtained through automated photoID studies. The state of the art for the automated photo identification of Risso’s dolphins is the algorithm SPIR (Smart Photo Identification of Risso’s dolphin, see Maglietta et al. Scientific Reports 2018), based on the analysis of dorsal fin scars. Images collected in the “Risso’s dolphin dataset” are part of the sightings data for the Risso’s dolphin collected from 2013 to 2019, during standardized vessel-based surveys carried out in the Gulf of Taranto, in the Northern Ionian Sea (North-eastern Central Mediterranean Sea), on board the catamarans of the Jonian Dolphin Conservation NGO. The study area covers a very complex topography zone of approximately 14,000 km2, spanning from Santa Maria di Leuca to Punta Alice. A narrow continental shelf, shaped by a steep slope and several channels, characterizes the western sector, while the eastern sector shows descending terraces towards the “Taranto Valley”, a NW-SE submarine canyon with no clear bathymetric connection to a major river system.All images were taken using a Nikon D3300 camera with a Nikon AF-P Nikkor 70–300 mm, f4,5–6,3 G ED lens. To avoid potential interference in dolphin behavior due to the presence of the vessel, sampling was interrupted by changing direction when specimens were observed at less than around 50 m. Moreover, all observers maintained a safe distance of no less than 5 m, while lowering speed or interrupting navigation to prevent collisions or possible injuries.

照片识别（photo identification，简称photoID）是一种非侵入式技术，旨在通过照片对单个动物进行个体识别，其核心假设为每一个体都具备可用于识别的独特特征。该技术尤其适用于研究高度移动且难以被观测的海洋物种，例如鲸类（Cetacea）。这类动物在海洋生物多样性保护中发挥关键作用：作为食物网中的顶级捕食者，它们维系着海洋生态系统的稳定与健康。通过照片识别研究获取的信息，可用于获取目标物种的种群丰度估算、社会动态、迁移模式以及位点保真度等方面的新知识。由于人工照片识别耗时耗力，且在处理大规模数据集时并不现实，因此采用先进的自动化技术能够为研究人员提供辅助，并加快个体照片识别的流程。开放数据集的公开可获取性，例如本次发布的题为"灰海豚数据集（Risso’s dolphin dataset）"的数据集，能够助力全自动化鲸类照片识别技术的开发与落地——既可以作为训练数据集使用，也可以提供已完成识别的个体用于后续匹配任务。总体而言，该数据集还可用于基于图像处理、计算机视觉与机器学习的新型方法开发，这些方法专注于图像裁剪、分割与识别任务。就本数据集的研究对象而言，灰海豚（Grampus griseus，Cuvier, 1812）是全球范围内认知程度最低的鲸类物种之一，其地中海种群被IUCN红色名录列为濒危（Endangered）等级。灰海豚的背鳍上带有可长期识别的天然斑纹，这些特征使得该物种尤其适配照片识别算法。因此，为填补该物种认知上的空白，自动化照片识别研究是关键的研究手段之一。目前用于灰海豚自动化照片识别的主流算法为SPIR（Smart Photo Identification of Risso’s dolphin，详见Maglietta等人发表于《Scientific Reports》2018年的研究），该算法基于背鳍疤痕特征开展分析。"灰海豚数据集（Risso’s dolphin dataset）"中收录的图像，源自2013年至2019年间在爱奥尼亚海北部（地中海东北部中部）塔兰托湾开展的标准化船舶调查所获取的灰海豚目击数据，该调查由约尼海豚保护协会（Jonian Dolphin Conservation NGO）运营的双体船执行。本次调查的研究区域范围约14000平方千米，涵盖从圣玛丽亚迪莱乌卡（Santa Maria di Leuca）到蓬塔爱丽丝（Punta Alice）的复杂地形海域。研究区域西部以狭窄大陆架为主要特征，该大陆架受陡峭斜坡与多条水道塑造；东部则呈现向"塔兰托谷"延伸的阶梯状下降地形——塔兰托谷为一条西北-东南走向的海底峡谷，其地形与任何大型河流系统均无明确的水深关联。所有图像均采用尼康D3300相机搭配尼康AF-P Nikkor 70–300mm f/4.5–6.3 G ED镜头拍摄。为避免船只对海豚行为造成潜在干扰，当观测到个体距离船只不足约50米时，研究人员会通过改变航向终止采样。此外，所有观测人员均保持至少5米的安全距离，并通过降低航速或暂停航行来避免船只与海豚发生碰撞或造成伤害。