Postal operators continue to experience increases in the volume of cross-border e-commerce parcels and this shows no sign of reducing – cross-border growth globally is set to have grown threefold by 2020, since 2015.
These items have complex international labels, and conventional OCR systems have difficulty identifying the right information. With the European Customs 2020 initiative, all European posts will be faced with the need to extract key information from CN22 and CN23 labels, including contents, value and weight, to allow recovery of VAT. If automated sorting systems cannot easily extract this information, it leads to significant manual handling costs and slow processing.
Lockheed Martin UK has developed a machine learning application called Minerva, which can quickly and effectively be taught to find this information on complex labels, without relying on restrictive and less effective techniques that require form identification.
Minerva can: improve automated sortation by finding delivery addresses, extract key customs information from CN22 and CN23 labels for VAT recovery, and reduce video coding workload and costs.
Importantly, Minerva can then pass this information to any existing machine OCR system, or be provided with its own OCR solution, giving the postal operator flexibility. This option of installing Minerva with an existing OCR solution can improve reading capability without the need to procure an expensive, new OCR solution.
Minerva has been in successful operational use in the PostNord Sweden parcel sorting operation since early 2018 and has demonstrated an immediate and significant increase in sortation rates of cross-border packets in live operations, increasing efficiency, reducing costs, reducing manual handling and speeding up the operation.