FoodML: Development of a food quality and safety risk management system, using cloud computing, big data and data science

目前，食品质量和安全控制在很大程度上依赖于监管检查和抽样制度。这些方法通常基于传统的化学和微生物分析，使得100%实时检测的最终目标在技术上、财务上和后勤上都是不可能的。（如振动光谱、高光谱/多光谱成像）作为快速有效的评估食品质量、安全和认证的方法开始受到欢迎;作为昂贵且耗时的传统微生物学技术的明智替代方案。由于此类分析产生的数据具有多维性，在解释结果之前，需要将输出与合适的统计方法或机器学习算法结合起来。虽然这些平台已经显示出巨大的潜力，以准确和定量地评估新鲜度概况（Panagou、Mohareb等人，2011年）（Mohareb，Iriondo等人，2015）和安全参数以及掺假（Ropodi，Panagou等人，2016），他们对先进的数据挖掘和统计算法的依赖是他们在食品生产中实际实施所面临的主要挑战，为了克服这些挑战，我们开发了sorfML（http：//elvis.misc.cranfield.ac.uk/SORF），这是一个与来自5个仪器平台的输出兼容的Web平台原型（见图），它提供了交互式数据可视化、多变量分析（主成分分析和层次聚类）以及使用存储的数据集开发预测模型以估计食品质量的能力。目前，该平台为用户提供了将实验数据集上传到服务器的方法，这要归功于真正通用的MongoDB NoSQL数据库后端，并开发分类和回归模型来估计质量参数。该博士学位的目的是将现有的sorfML平台扩展为一个支持云的框架，支持整个生产链中食品的实时监控。为达致这个目标，我们会采用一系列先进的便携式传感器，以检验它们是否适合作为“连接装置”，预测各种易腐食品的质量和安全指数。一系列机器学习和模式识别模型将被开发并集成到云系统中。这包括普通最小二乘，逐步线性分类和回归，主成分回归，偏最小二乘判别分析，支持向量机，随机森林和k-最近邻。

Currently, food quality and safety controls relies heavily on regulatory inspection and sampling regimes. Such approaches are often based on conventional chemical and microbiological analysis, making the ultimate goal of 100% real-time inspection technically, financially and logistically impossible.Over the past decade, rapid non-invasive techniques (e.g. vibrational spectroscopy, hyperspectral / Multispectral imagining) started gaining popularity as rapid and efficient methods for assessing food quality, safety and authentication; as a sensible alternative to the expensive and time-consuming conventional microbiological techniques.Due to the multi-dimensional nature of the data generated from such analyses, the output needs to be coupled with a suitable statistical approach or machine learning algorithms before the results can be interpreted. Although these platform has been showing great potentials to accurately and quantitatively assess freshness profiles (Panagou, Mohareb et al. 2011) (Mohareb, Iriondo et al. 2015) and safety parameters as well as adulteration (Ropodi, Panagou et al. 2016), their dependence on advanced data mining and statistical algorithms made was the main challenge facing their practical implementation across the food production and supply chain.In order to overcome these challenges, we have developed sorfML (http://elvis.misc.cranfield.ac.uk/SORF), a Web platform prototype compatible with outputs from 5 instrumental platforms (See Figure) which provides means for interactive data visualisation, multivariate analysis (Principal component analysis and hierarchical clustering), as well as the ability to use stored datasets to develop predictive models to estimate food quality. Currently, the platform provides users with means to upload their experimental datasets to the server, thanks to the truly generic MongoDB NoSQL database backend, and to develop classification and regression models to estimate quality parameters.The aim of this PhD is to expand the existing sorfML platform into a cloud-enabled framework that supports real-time monitoring of food products throughout the production chain. In order to achieve this, a series of advanced portable sensory devices will be deployed to examine their suitability as "Connected devices" in predicting quality and safety indices for various food perishable food products. A series of machine learning and pattern recognition models will be developed and integrated within the cloud system. This includes Ordinary Least Squares, Stepwise Linear classification and regression, Principal Component regression, Partial Least Squares discriminant analysis, support vector machine, Random forests and k-Nearest Neighbours.