b'Partner News Autumn 2023 11AcademicEmbedding AI in the next generation of barcode scanners By Dr Zoe Jeffrey CEng MIET and Prof Yichuang Sun (University of Hertfordshire), Neil Benn, Jini Thomas, Mahdi Sayadi, Alex Beasley CEng MIET and Paul Withers (Ziath Ltd.)This project is a knowledge transfer2D BarcodeThe artificial intelligence technique Scanner with partnership collaboration between thesample tubes employed in this work used Deep Neural University of Hertfordshire and Ziath Ltd,Networks (DNNs). The DNNs role in sponsored by Innovate UK. Ziath Ltd.the new application was to extract 2D specialises in instrumentation controldatamatrix barcodes from sample tube and information management in bothimages. The DNN algorithm was designed the academic and the pharmaceutical/ to be very lightweight allowing for quick biotech industry sectors. Its focus is onexecution and the results showed a the application of laboratory automation,substantial improvement when compared to specifically the management of large samplethe pre-existing heuristic methods, the DNN libraries using 2D data matrix barcodes onapproach was almost ten times faster to sample tubes. execute with virtually 100% accuracy. Datamatrix barcodes play a key role inThe DNN is implemented on embedded Portable tracking and tracing both biological and2D Barcodetechnology, in this instance a Field compound samples. These barcodes areScanner withProgrammable Gate Array (FPGA). usually lasered onto the underside ofembedded AIFPGAs allow for custom circuity to be sample tubes, and the tubes are stored intechnology created. Due to the custom nature of the racks. Barcode reading is conducted usingimplementation this yields a very high-a barcode reader that scans the bottom ofspeed DNN, faster than can be achieved a rack of tubes and decodes all barcodes inon a standard PC processor. The inclusion one go. This process is efficient in theory,of the FPGA within the system opens however, in reality, there are regular issuesnew possibilities for the way in which the with identifying the barcodes on the bottombarcode scanners can be implemented. of the tubes.The power of the embedded FPGA makes possible to build a stand-alone mobile Often, inadequate ambient lighting,scanner, capable of decoding an entire background image noise, variation inrack (96 tubes) in a sub-second timeframe lasering and poor material quality can yieldwhile having low power requirements and tube barcode images that are too difficultoutperforming a traditional high-spec to detect with traditional machine visionlaptop or desktop PC.techniques. However, a human can always resolve these barcodes, even in adverseThe outcome of this project is Ziaths conditions. Therefore, it is reasonable tolaunch of the first truly portable, real-time, assume that artificial intelligence techniques2D barcode scanner with embedded AI can be employed to increase the successtechnologies that can be used in the lab as rate for identifying datamatrix barcodes. 2D barcode scanner with embedded AI technologies - GUI well as in front of freezers or sample stores.'