Bridging ecology and technology: Using citizen science and artificial intelligence to track fish health

关于鱼类分布及其疾病状况的数据对野生渔业的养护和管理至关重要。垂钓者报告说，福利问题的负担越来越重，特别是那些会导致皮肤损伤的突出皮肤病原体，而且往往是一般健康的关键指标。腐竹属物种的卵菌尤其有问题；随着鱼类在气候变化的影响下变得更加紧张和免疫功能受损，其流行率不断上升，由此导致的鱼皮肤上的真菌样生长可以用肉眼看到。在英国，野生渔业数据是由威尔士自然资源局(NRW)和环境局(EA)等政府机构使用耗时、费力和昂贵的方法收集的。因此，关于可见鱼类疾病的分布和范围的数据稀少，导致即使是常见物种的知识也存在很大差距。然而，通过在社交媒体上分享的图像提供的公开贡献的数据基本上是未开发的来源，可以通过提供关于鱼类分布和疾病状况的广泛空间数据来填补这一知识空白。社交网络拥有超过20亿的用户基础，反映了25%的人类人口的观察到的现实。使用人工智能机器学习技术，从提交给社交媒体的图像中，我们将开发一个自动监测系统(AMS)，以识别英国各地的鱼类种类、它们的位置和疾病迹象。作为定向和结构化调查的替代方案，我们将探索社交媒体内容的价值，公众可以在这些内容中临时拍摄和描述野生动物，而不是特定的科学意图。这些用智能手机拍摄的图像带有地理标记，因此可以提供丰富的经过验证的野生动物分布数据来源，构成了我们所说的“被动公民科学”。我们将对英国各地野生淡水鱼测绘的社交媒体观测质量进行系统评估，以使用在线照片共享网站、Flickr、Twitter等上的文本标签和图片内容预测物种分布数据。我们将使用机器学习和计算机视觉过程，识别鱼类物种并量化鱼类上任何可见的疾病(例如，腐臭)。这些数据将使我们能够绘制野生鱼类的分布地图，同时绘制它们的疾病状态图，从而使我们能够快速识别疾病分布。作为第二个目标，我们将发起一个专门的公民科学项目，深入绘制威尔士河流上的鱼类分布和健康地图。与利益相关者‘怀伊和乌斯克基金会’合作，我们将使用他们的‘捕鱼护照’推出一个公民科学项目。Wye和Usk上的垂钓者需要注册一日通行证，因此提供了一个基础设施，我们可以通过这个基础设施向垂钓者提供如何将图像上传到照片共享网站Flickr的说明。从这一非常低的公民科学努力中，我们将提取位置数据，并使用机器学习工具来识别图像中的物种以及任何疾病的存在和程度。我们将评估这项技术作为该集水区疾病的实时报告。

Crucial to conservation and management of wild fisheries are data on fish distribution and their disease status. Anglers have reported an increasing burden of welfare issues, particularly for those prominent skin pathogens which cause skin lesions, and are often a key indicator of general health. Oomycetes of Saprolegnia species are particularly problematic; increasing in prevalence as fish become more stressed and immunocompromised in the light of climate change, and the resulting fungal-like growths on the skin of the fish are visible by eye. In the UK, wild fisheries data are collected via government agencies, such as Natural Resources Wales (NRW) and the Environment Agency (EA) using methods that can be timeconsuming, laborious, and expensive. Data on the distribution and extent of visible fish diseases are, as such, sparse, leading to large gaps in knowledge even for common species. There exists, however largely untapped sources of publicly contributed data via images shared on social media that could fill this knowledge gap by providing extensive spatial data on fish distribution and disease status. Social networks have a user base of over 2 billion, reflecting the observed reality of 25% of the human population. Using AI machine learning techniques from images submitted to social media we will develop an automatic monitoring systems (AMS) to identify, across the UK, fish species, their location and signs of disease.As an alternative to directed and structured surveys, we will explore the value of social media content in which members of the public photograph and describe wildlife ad hoc, and with no particular scientific intent. Such images, taken with smart phones are geo-tagged and so could provide a rich source of validated wildlife distribution data, constituting what we term 'passive citizen science'. We will conduct a systematic evaluation of the quality of social media observations for wild freshwater fish mapping across the UK to predict species distribution data using the textual tags and image content of photos on online photo sharing sites, Flickr, Twitter, etc. Using machine learning and computer vision processes, we will identify fish species and quantify any visible disease on the fish (e.g. Saprolegnia). Such data will allow us to map the distribution of wild fish whilst simultaneously mapping their disease status, so allowing us to rapidly identify disease distribution.As a second objective we will initiate a dedicated citizen science project to map fish distribution and health intensively on Welsh rivers. Working with stakeholders 'Wye and Usk Foundation' we will use their 'fishing passport' to roll out a citizen science project. Anglers on the Wye and Usk are required to register for a day pass so providing an infrastructure via which we can provide instructions to angler on how to upload images to the photo-sharing website Flickr. From this very low citizen science effort we will extract location data and use the machine learning tools to identify both the species in the image and the presence and extent of any disease. We will evaluate this technique as a real-time reporter of disease on this catchment.