Automated corrosion detection in Oddy test coupons using convolutional neural networks

Abstract The Oddy test is an accelerated ageing used to determine whether a material appropriate for the storage, transport, or display of museum objects. levels corrosion seen on coupons silver, copper, and lead indicate material’s safety use. Although conducted in heritage institutions around world, it often critiqued lack repeatability. Determining level manual subjective process, which outcomes are affected by differences individuals’ perceptions practices. This paper proposes that more objective evaluation can be obtained utilising convolutional neural network (CNN) locate metal classify their levels. Images provided Metropolitan Museum Art (the Met) were labelled object detection train CNN. CNN correctly identified type 98% set Met’s images. also collected from American Institute Conservation’s wiki page. These images suffered low image quality missing classification information needed Experts cultural evaluated images, but there was high disagreement between expert classifications. Therefore, these not However, proved useful testing limitations trained data when applied different protocols photo documentation procedures. presents effectiveness Met non-Met coupons. Finally, this next steps produce universal CNN-based tool. Graphic