Plastic packaging, a complete recognition and monitoring system based on AI and fusing RGB-based computer vision with Near Infrared spectral Imaging (NIR SI)

到2040年，全球将产生33亿吨垃圾，其中塑料垃圾将翻一番，达到4.3亿吨。据预测，由于缺乏数据，低效，不准确的（人工）审计和处理，从而抑制了废物的有效分类，因此只有13%的废物将被回收利用。目前，使用近红外（NIR）相机的分拣机提供了部分解决方案；然而，它们非常昂贵（每单位20万至30万英镑），而且技术上有限（近红外视觉无法区分所有类型的废物），这意味着塑料包装的回收/再循环率明显低于目前的英国塑料公约目标（2019年约为50%，WRAP-2020对70%的目标）。为了应对这一挑战，灰鹦鹉公司（GP）与蓝绿视觉公司（BGV）合作开发了一种混合废物识别和监测系统，以补充和增强近红外分类机，并以整体方法解决缺乏准确数据的问题。该系统将灰鹦鹉公司基于人工智能的RGB相机（红绿蓝即人类视觉）技术与BGV公司的高速点对点近红外系统（比目前的高光谱近红外相机便宜5-10倍）相结合，提供大规模的自动化、经济高效、准确的塑料废物识别（可转移到所有废物类型）。至关重要的是，该系统将能够识别所有类型的材料和包装，例如黑色塑料或低密度聚乙烯（LDPE）套管的聚对苯二甲酸乙二醇酯（PET）瓶，现有的近红外系统目前无法识别/分类。通过提供改进工厂流程所需的准确和实时数据，并整合拟议的系统来增强分拣机器，合作伙伴预计塑料回收率将保守地提高5%；为实现英国塑料公约的第三个目标做出了重大贡献。独特的废物识别和监测系统将使废物公司能够准确和全面地大规模审计废物，这对于满足将于2023年生效的严格的扩展生产者责任（EPR）法规至关重要，（该法规将要求每8吨取样一次，而目前每125吨取样一次）提供更大的问责制和更高的回收率。这个项目将成为提供这种服务的催化剂。此外，该系统将支持对整个工厂进行全球优化的新方法，而不是对单个分拣机进行局部优化，从而提高所有废物类别的回收率，并有助于建立更强大的英国回收系统。

By 2040, c.3billion tons of waste will be produced globally, with plastic waste doubling to 430million tons. It is predicted that only 13% will be recycled due to inefficient, inaccurate (manual) auditing and processing, driven by a lack of data, thus inhibiting the effective sorting of waste. Sorters using Near InfraRed (NIR) cameras, in place today, provide a partial solution; however they are extremely expensive (£200k-£300k/unit), and technically limited (NIR vision is unable to differentiate all types of waste), meaning plastics packaging recovery/recycling rates are significantly lower than current UK Plastics Pact targets (c.50% in 2019, WRAP-2020 vs. 70%-target).To address the challenge, Greyparrot (GP) in collaboration with Blue Green Vision (BGV) are developing a hybrid waste recognition and monitoring system to complement and enhance NIR sorting machines, and tackle the lack of accurate data with a holistic approach. The system will combine Greyparrot's AI-based technology using RGB cameras (Red-Green-Blue i.e.human vision) with BGV's high-speed point-to-point NIR system (5-10x cheaper than current hyperspectral NIR cameras) to provide automated, cost-effective, accurate recognition of plastics waste (with transferability to all waste types) at-scale. Critically, the system will enable the identification of all types of material and packaging, e.g including black plastics or Polyethylene terephthalate (PET) bottles with a Low-Density Polyethylene (LDPE) sleeve, not currently recognised/sorted by existing NIR systems. By providing accurate and real-time data required to improve processes in the plant, and integrating the proposed system to enhance sorting machines, the partners anticipate a conservative 5% increase in plastics recovery rates; making a significant contribution towards delivering the 3rd target of the UK Plastics Pact.The unique waste recognition and monitoring system will enable waste companies to accurately and comprehensively audit waste at-scale, critical to meeting strict Extended Producer Responsibility (EPR) legislation set to take effect in 2023,(with which samples will be required as often as every 8 tonnes against every 125 tonnes currently) providing greater accountability and increased recycling rates. This project will be a catalyst to deliver such a service. In addition, the system will support new approaches to carry out a global optimisation of entire plants instead of local optimisations in singular sorters, leading to better recovery rates in all waste categories and helping to build a stronger UK recycling system.