CryptoBioVision: Applying Computer Vision for Cryptic Species Discovery

地球上的大部分生物多样性仍未被描述、研究和观察。这有时被称为“林奈短缺”，以伟大的瑞典博物学家卡尔·林奈的名字命名。要自信地研究和回答生态学和进化中许多最基本的问题，例如：现代生物多样性是如何以及为什么进化的？对现代物种多样性、它们彼此之间的关系、它们如何生活以及在哪里生活，有一个坚实的了解是必要的。但也许更紧迫的是，林奈短缺从根本上阻碍了我们对全球生物多样性危机的严重性的理解，以及最终如何应对它。造成短缺的最大原因之一是，在科学家眼中，许多近亲物种看起来非常相似或难以区分，因此被称为“隐种”。因此，为了帮助弥补这一缺口，我们需要开发新的方法来加速发现和帮助描述隐藏物种。在我们的提案中，我们的目标是表明计算机视觉（以下简称CV）技术的最新进展可以帮助提供解决方案。CV是一个快速发展的领域，计算机被训练来识别、提取和测量数字图像中的信息。虽然CV有望成为生态进化和分类学研究的重要工具，但目前还没有研究表明CV可以用于隐藏物种的发现和描述。我们的总体目标是将新兴技术（CV）与传统和成熟的方法（田野调查，DNA条形码，分类学）结合起来，以加强和加速生物多样性的发现，并帮助描述来自多种栖息地类型的隐物种。我们的试点研究表明，我们可以训练我们的模型，使用数字图像将单个标本与正确的物种相匹配。我们还表明，CV可以识别出区分对隐种的形态特征。这些新颖和创新的方法将为科学家提供加速发现和描述隐藏物种的手段，并更好地研究它们的生态和进化。我们开发了三个概念验证研究，将CV应用于现场收集和自然历史博物馆标本的数字图像。我们重点研究了温带、亚热带和热带地区的沿海、深海和陆地生境中具有高度隐蔽多样性的动物类群。在第一项研究中，我们建议进行实地考察，从墨西哥巴哈半岛（PI在那里有很强的研究背景）这一高度隐蔽多样性的地区收集沿海软体动物标本。我们将应用我们的CV模型来区分Lottia属帽贝的隐种。试点结果是有希望的，并揭示了外壳的外边缘是有用的区分特征。在第二项研究中，我们将把我们的CV模型应用于600个深海片脚类（Eurythenes属）标本，该物种具有较高的隐多样性。试验结果表明，我们的CV模型将大约90%的标本识别为正确的物种，并突出了背部区域作为独特的兴趣区域。在我们的第三项研究中，我们与我们的NMH合作伙伴合作开发CV模型，用于识别数字固定昆虫收藏中的隐物种，这是同类中最大和最多样化的收藏（来自全球各地的数百万个标本）。在这个全球变化的时代，如果我们要：(1)充分掌握生物多样性危机的严重性，(2)对此采取措施，我们需要开发突破性和创新的方法来帮助弥合“林奈短缺”。总的来说，我们的研究旨在首次展示将CV与传统方法结合起来解决林奈不足的强大效用。我们未来的目标是开发一个在线平台（CryptoBioVision），允许来自全球的研究人员将我们的CV模型应用于他们的研究小组。

Most of the biodiversity on the planet remains undescribed, unstudied, and unseen. This is sometimes referred to as the "Linnean shortfall", named after the great Swedish naturalist Carl Linneaus. A solid understanding of modern species diversity, their relationships to each other, how they live and where, is necessary to confidently study and answer many of the most basic questions in ecology and evolution, such as: how and why did modern biodiversity evolve? But perhaps more urgently, the Linnean shortfall fundamentally impedes our understanding of the magnitude of the global biodiversity crisis, and ultimately, how to respond to it.One of the biggest reasons for the shortfall is because many closely related species look very similar or indistinguishable to the eyes of scientists, so called "cryptic species". Thus, to help close the shortfall, we need to develop new methods to accelerate the discovery and aid the descriptions of cryptic species. In our proposal, we aim to show that recent advances in Computer Vision (hereafter CV) technology can help provide a solution. CV is a rapidly developing field in which computers are trained to recognise, extract and measure information from digital images. While CV is poised to become an essential tool for eco-evolution and taxonomy research, there are currently no studies showing that CV can be used for cryptic species discovery and description. Our overarching aim for this proposal is to bring together emerging technologies (CV) with traditional and established methods (fieldwork, DNA barcoding, taxonomy) to enhance and accelerate biodiversity discovery and aid descriptions of cryptic species from multiple habitat types. Our pilot research shows that we can train our models to match individual specimens to the correct species using digital images. We also show that CV can identify the morphological features that distinguish between pairs of cryptic species. These novel and innovative methods will provide scientists the means to accelerate the discovery and descriptions of cryptic species and to better study their ecology and evolution.We develop three proof-of-concept studies that apply CV to digital images of field collected and natural history museum specimens. We focus on animal groups with high cryptic diversity from coastal, deep-sea, and terrestrial habitats - spanning temperate, subtropical, and tropical regions. In the first study, we propose field trips to collect specimens of coastal molluscs from a region of high cryptic diversity, the Baja peninsula in Mexico (where the PI has a strong research background). We will apply our CV models to distinguish between cryptic species of limpets in the genus Lottia. Pilot results are promising and reveal that the outer margins of the shells are useful distinguishing features. In the second study, we will apply our CV models to >600 specimens of deep-sea amphipods in the genus Eurythenes, which has high cryptic diversity. Pilot results show that our CV models identify >90% of specimens to the correct species and highlights dorsal regions as distinctive areas of interest. In our third study, we work with our NMH partners to develop CV models for identifying cryptic species within the digital pinned insect collection, which is the largest and most diverse collection of its kind (millions of specimens from all over the globe). In this era of global change, we need to develop ground-breaking and innovative approaches to help close the "Linnean shortfall" if we are to: (i) fully grasp the magnitude of the biodiversity crisis and (ii) do something about it. Collectively, our studies aim to be the first to showcase the powerful utility of integrating CV with traditional methods for addressing the Linnean shortfall. Our future goal is to develop an online platform (CryptoBioVision) that will allow researchers from across the globe to apply our CV models to their study groups.