Runoff: Remote worming - developing computer learning for high throughput identification of earthworm populations as an indicator of soil health

近年来，对土壤健康和农业管理在促进土壤可持续性方面的作用的了解有所增加，特别是在“改善土壤健康”被列入英国政府的25年环境计划之后。英国退欧后的农业补贴可能会用于改善环境，因此现在有必要开发监控系统。蚯蚓可以被描述为土壤健康的象征，推动养分循环和水分渗透过程-如果土壤中有丰富的蚯蚓种群，土壤动物的其余部分也将健康，土壤化学和土壤结构也是如此。传统上，蚯蚓种群监测费时费力，而且可能不准确，由于评估员的能力，公民科学监测项目已经进行了试验，以降低成本，但尚未在全国推广。利用计算机学习作为一种在田间范围内高通量识别蚯蚓丰度的工具，可以在英国境内的农田中实施，以提供对蚯蚓活动的当前评估。由于蚯蚓挖洞减少了径流并改善了土壤孔隙度，这种方法提供了一种低成本、快速的监测评估工具，将提供一种“生物健康评估”，可以向农民提供信息和教育，并导致农业管理的改善。该方案旨在开发一种深度学习算法工具来高通量地对蚯蚓铸件进行现场检测和计数。如果成功，基于这种生物图像分析的软件可以通过智能手机应用程序或无人驾驶车辆部署，从而在全国范围内监测蚯蚓。到目前为止，已经开发了许多用于测量土壤/土壤健康的应用程序，但没有一个将计算机深度学习的对象识别与蚯蚓活动相结合，这是一个明显的研究空白，本提案旨在填补这一空白。

Understanding soil health and the effect agricultural management has in promoting the sustainability of the soil has increased in scrutiny in recent years particularly since "improving soil health" was included in the UK Government's 25 year plan for the environment. Post-Brexit farming subsidies are likely to be given for environmental improvement, therefore there is a need to develop monitoring systems now. Earthworms can be described as the emblem of a soil health, driving nutrient cycling and water infiltration processes - if there is an abundant earthworm population within the soil, the likelihood is the rest of the soil fauna will also be healthy as will the soil chemistry and soil structure. Traditionally earthworm population monitoring is laborious and can be inaccurate, due to the ability of the assessor, citizen science monitoring programs have been trialled to reduce costs, but have not been extended across the country. Utilising computer learning as a tool for high throughput identification of earthworm abundance at a field-scale could be implemented across farmland within the UK, to provide a current assessment of earthworm activity. As earthworms burrowing reduces water runoff and improves soil porosity, this method provides a low cost, fast monitoring assessment tool that would provide a "biological health assessment" that could inform and educate farmers and lead to improvements in agricultural management. This proposal aims to develop a deep learning algorithm tool to detect and count earthworm casts in-situ at high-throughput. If successful, software based on this bioimage analysis could be deployed via smartphone app or unmanned vehicle, leading to monitoring of earthworms nationally at the field-scale. To date there have been many apps developed to measure soil / soil health, but none combine computer deep-learning for object recognition with earthworm activity, this is a clear research gap, that this proposal aims to fill.