Precision agriculture: AI- and Expert- based approach to forcast fruit production in high intra-field variation settings

预测软果产量的一个重大挑战是田间作物差异大，即田间特定部分的表现比其他部分差。大多数预测模型的运作方式是预测整个作物的平均生长率，然而，对任何显示变化的作物的产量进行外推可能会导致预测错误。这是因为作物生长速率与环境变量之间的关系非常复杂，具有多重相互作用（例如温度x辐射x水）和非线性（光合作用对光的响应）。然而，随着大数据分析和人工智能的爆发，这些信息可用于开发算法和模型，这些算法和模型对农民来说是有用且廉价的工具。鉴于该行业的高度竞争性，种植者不仅需要用于监测田间表现的工具，还需要尽可能准确地预测产量的工具。目前，种植者深入研究基于经验/经验的预测，由于他们未能考虑的因素/模式，这些预测有时并不代表现实。这意味着他们可能会向零售商提供错误的估计，导致产品浪费或无法满足需求。博士生将与水果行业密切合作，使用大量的数值数据在云端开发定制的人工智能和深度学习技术（最大限度地降低昂贵设备的成本），这些技术将利用重要的先验知识，如季节变化，并将根据后验信息（如天气预报或其他相关信息）为未来预测提供信息。这些模型将纳入基于专家的投入-而不仅仅是基于数据的投入-以便为系统增加专家知识和/或制约因素。它还将包括一个分层和/或基于规则的组件，以考虑场内变化，从而分别为场内的每个部分提供预测。在农民的帮助下，将绘制该领域的初步地图。正在进行的CTP博士研究的其他输入可能包括使用机器人进行视频捕获，如果需要，可以将其用作开发模型的额外模式。博士生将开发跨越机器学习，农业和工业需求的各种技能，受益于高度跨学科的团队。1：Georgios Leonardo博士和Stefanos Kollias教授在林肯大学计算机科学学院工作。两人都是机器学习和人工智能方面的专家，参与了与农业、能源和工业相关的各种资助项目。两人都有博士监督的证明记录。Kollias教授指导完成了超过43名博士生。2：大卫梅先生在林肯大学的林肯农业食品技术研究所工作。他在农业食品领域拥有丰富的经验，对与此相关的行业和学术界的需求和需求有着深刻的了解。

A significant challenge in forecasting soft fruit yield is high intra-field crop variation, i.e. specific parts of the field performing worse than others. Most forecast models operate by predicting the mean growth rate as a crop as a whole, however, extrapolating yield over any crops showing variation is likely to cause forecast errors. This is because crop growth rate to environmental variables is highly complex, has multiple interactions (e.g. temperature x radiation x water) and non-linear (photosynthesis response to light). However, with the outburst of big-data analytics and AI this information can be used to develop algorithms and models that can be useful and inexpensive tools to the farmers. Given the highly competitive nature of the sector, growers need tools not only for monitoring the field performance but also for providing them with as accurate yield predictions as possible. Currently, growers delve into experience-based/empirical predictions, which sometimes do not represent the reality due to factors/patterns they fail to consider. That means they might provide wrong estimates to the retailers, leading to either wasting products or missing to meet the demand. The PhD student will work closely with the fruit industry to use large amounts of numerical data to develop bespoke AI and deep learning techniques on the cloud (minimising the cost of expensive equipment) that will leverage important a priori knowledge, such as seasonal variations, and will inform future predictions based on posterior information, such as weather forecast or other relevant information. The models will incorporate expert-based input - not just data-based one - to add the expert knowledge and/or constraints to the system. It will also include a hierarchical and/or rule-based component to consider the intra-field variations, hence providing predictions for each section of the field separately. Initial maps of the field will be created with the help of the farmers. Additional input from ongoing CTP PhD studies might include video capture using robots that can be used as an extra modality to the developed model, if needed. The PhD student will develop various skills spanning machine learning, agriculture and industrial demands, benefitting by the highly interdisciplinary team.1: Dr Georgios Leontidis and Professor Stefanos Kollias are with the School of Computer Science at the University of Lincoln. Both are experts on machine learning and AI, participating in various funded projects related to agriculture, energy and industry. Both have a proven record of PhD supervision. Professor Kollias has supervised to completion more than 43 PhD students.2: Mr David May are with the Lincoln Institute for Agri-food technology at the University of Lincoln. He has vast experience of the agri-food sector with deep knowledge of the demands and needs of industry and academia related to this.