Hyuk Jun Yoo1,2,Leslie Ching Ow Tiong1,Na Yeon Kim1,2,Kwan-Young Lee2,Sang Soo Han1,Donghun Kim1
Korea Institute of Science and Technology1,Korea University2
Hyuk Jun Yoo1,2,Leslie Ching Ow Tiong1,Na Yeon Kim1,2,Kwan-Young Lee2,Sang Soo Han1,Donghun Kim1
Korea Institute of Science and Technology1,Korea University2
Robot-based automation methods for material synthesis have recently garnered much attention because they can substantially accelerate the material development process. Some recent examples were reported, which involves organic or inorganic material synthesis in energy applications (catalysis and photovoltaics) [1-3]. Despite the substantial promise of these methods, surveillance-free environments may lead to dangerous accidents primarily due to hardware control errors. Object detection techniques can play important roles in addressing these safety issues; however, state-of-the-art detectors, including single-shot detector (SSD) models, suffer from insufficient accuracy in environments involving complex and noisy scenes due to uneven visual environments. With the purpose of improving safety in a surveillance-free laboratory, we developed the deep learning (DL)-based object detector, namely, densely connected single-shot detector (DenseSSD) with a densely connected mechanism. For the foremost and frequent problem of detecting vial positions, DenseSSD achieved a mean average precision (mAP) over 95% based on a complex dataset involving both empty and solution-filled vials, greatly exceeding those of conventional detectors; such high precision is critical to minimizing failure-induced accidents. Additionally, DenseSSD was observed to be highly insensitive to the environmental changes, maintaining its high precision under the variations of solution colors or testing view angles. The roughness of DenseSSD would allow the utilized vision module settings to be more flexible. This study verified that DenseSSD is practical for enhancing safety in an automated material synthesis environment, and it can be extended to diverse applications where high detection accuracy and speed are both needed.<br/><br/><i><u>[1] Nature</u></i><u> <b>538</b>, 237-241 (2020)</u><br/><i><u>[2] Nature</u></i><u> <b>559</b>, 377-381 (2018)</u><br/><i><u>[3] Science</u></i><u> <b>365</b>:eaax1566 (2019)</u>.