Accelerating the transition towards a circular economy by using machine vision and digital platforms

通过使用机器视觉和数字平台，加速向循环经济的过渡，实现更智能的特征描述、无处不在的跟踪和自动化的垃圾分类。先前的研究已经开发出一种名为“Replecleye”的概念验证神经网络，它提供了一个经济实惠的系统：识别和分类可回收物品，提高分拣效率，减少人工劳动，推动垃圾数据革命。该项目将回答以下问题：该系统在工业条件下或在更多可能的分类下如何运行？神经网络架构是否可以改进，例如，当看到一个比萨饼盒时，系统可以100%准确地理解它是一个纸箱，尽管它可能被另一个物品撕裂，半隐藏在另一个物品中，并且覆盖着污垢？使用无监督学习（AI）技术是否能够对废物流中的所有物品进行绘图和分类？如果设备的成本可以降低，我们是否可以开始跟踪单个废物，以创建与供应链一样高效的清除链？在对成千上万的废弃物进行品牌级分类时，是否仍能保持准确，从而消除对条形码及其相关视线问题的需求？强化学习（AI）可以用来创建机器人垃圾拾取器吗？这项技术能否使生产者责任延伸计划走向个性化？这项工作将与几个工业伙伴完成，包括Biffa（一家材料回收设施运营商）、Bywaters（一家材料回收设施运营商）、VITO（一家比利时研究机构）和布鲁塞尔机场。

Accelerating the transition towards a circular economy by using machine vision and digital platforms to enable smarter characterisation, ubiquitous tracking and automated sorting of waste.Previous research has developed a proof of concept neural network called Recycleye which provides an affordable system to: identify and classify recyclables, improve sorting efficiency, reduce human drudgery and drive a waste data revolution. This project will answer the following questions: How would the system perform under industrial conditions or with a greater number of possible classifications? Can the neural network architecture be improved such that for example when seeing a pizza box, the system understands with 100% accuracy that it is a carton despite the fact it may be torn, half hidden by another item and covered in dirt? Could the use of unsupervised learning (AI) techniques enable the mapping and classification of all items within a waste stream? If the cost of the device can be reduced, can we start tracking individual waste items to create removal chains as efficient as supply chains? Will brand level classification remain accurate when performed on thousands of waste items, thereby removing the need for barcodes and their associated line of sight issues? Can reinforcement learning (AI) be used to create robotic waste pickers? Can the technology allow extended producer responsibility schemes to move towards individualization? The work will be completed with several industrial partners including Biffa (a material recovery facility operator), Bywaters (a material recovery facility operator), VITO (a Belgian research institution) and Brussels Airport.