Abstract Photo identification (PID) in animal studies has been a widely used method for identifying individuals of many species based on unique natural markings and patterns. The use of PID has facilitated investigations in which residency, home ranges, and growth rates have been assessed. However, many PID studies in the past have relied heavily on manual photo matching. More recently, computer-assisted PID programs have been used to identify individuals of different sea turtle species, and reduced time investment in identifying individuals within specific populations. Still, some computer-based PID programs require significant time investment in ensuring photos are captured at consistent angles and lighting conditions, pre-processing image manipulations, and post-processing manual matching confirmation of potential matches provided by the program. For PID to be an effective time and money saving mechanism for wildlife research and conservation, these common drawbacks need to be addressed with a computer-assisted PID program that reduces manipulation and time investment burden, and consistently provides accurate and reliable results. In this study, we evaluated the accuracy of matching individual face images using the HotSpotter (HS) PID program by building a database of 2136 images of hawksbill (Eretmochelys imbricata) turtles, then querying the database with 158 new images to find matches for individual turtles. Overall, we found that with almost no pre-processing manipulation, and with images from highly variable underwater conditions, qualities, and angles, HS correctly matched individuals in the first choice 80% of the time, increasing to 91% in the first six choices. When assessing in-water images only, accuracy for matching increased from 84% in the first choice, to 94% by the sixth choice. We suggest that the integration of HS technology into a global, web-based PID system will increase the ability to remotely identify individual marine organisms on a global scale, and improve usability for community scientists who may have little to no technical training.

中文翻译：

---

HotSpotter：使用计算机驱动的照片 ID 应用程序来识别海龟

摘要 动物研究中的照片识别 (PID) 已成为一种广泛使用的方法，用于根据独特的自然标记和图案识别许多物种的个体。PID 的使用促进了对居住、家庭范围和增长率进行评估的调查。然而，过去的许多 PID 研究严重依赖于手动照片匹配。最近，计算机辅助 PID 程序已被用于识别不同海龟物种的个体，并减少了识别特定种群内个体的时间投入。尽管如此，一些基于计算机的 PID 程序需要投入大量时间来确保以一致的角度和光照条件捕获照片、预处理图像处理、和程序提供的潜在匹配的后处理手动匹配确认。为了使 PID 成为野生动物研究和保护的有效时间和金钱节省机制，需要通过计算机辅助 PID 程序来解决这些常见的缺点，以减少操作和时间投资负担，并始终如一地提供准确可靠的结果。在这项研究中，我们通过构建一个包含 2136 张玳瑁（Eretmochelys imbricata）海龟图像的数据库，然后使用 158 张新图像查询该数据库以查找单个海龟的匹配项，从而评估使用 HotSpotter (HS) PID 程序匹配单个人脸图像的准确性. 总的来说，我们发现几乎没有预处理操作，以及来自高度可变的水下条件、质量和角度的图像，HS 在第一选择中正确匹配个体的概率为 80%，在前六个选择中增加到 91%。仅评估水中图像时，匹配精度从第一选择的 84% 增加到第六选择的 94%。我们建议将 HS 技术集成到基于网络的全球 PID 系统中，将提高在全球范围内远程识别单个海洋生物的能力，并提高可能几乎没有技术培训的社区科学家的可用性。