Machine Learning Solution for Roman Provincial CoinsFully automated, scalable cloud based solution for curatorial activities for targeted 300K coins collection
Oxford University’s Gardens, Libraries, and Museums (GLAM) form one of the greatest concentrations of university collections in the world. Comprising over 21 million objects, specimens, and printed items, they constitute one of the largest and most important research repositories in the world and provide an outstanding resource for scholars, students, and members of the public.
Several departments in GLAM employ various systems and are pursuing or planning digitalization projects, but none are actively utilising AI/ML technology. All parties have shown a strong desire to use these technologies to achieve the University’s and GLAM’s strategic goals.
Oxford GLAM and cloudmantra focused on delivering technical proof of concept and proving/ disproving the suitability of Machine Leaning tooling with collections objects to enable and maximize mundane and usually very manual tasks, preservation, quality assurance, or creative search in digital collection enabling value co-creation at the curatorial level.
About University of Oxford’s Gardens, Libraries & Museums
The University of Oxford’s Gardens, Libraries & Museums house some of the world’s most significant collections. They are important places of scholarly enquiry and serve as the front door to the wealth of knowledge and research generated at Oxford, welcoming over 3 million visitors each year.
The Ashmolean Museum of Art and Archaeology, the Oxford University Museum of Natural History, the Pitt Rivers Museum, and the History of Science Museum are all part of GLAM. The GLAM team picked the Ashmolean Museum’s Roman Provincial Coinage collection as a viable well-cataloged resource that would serve as the prototyping option for optimising access to the rich collection for digital teaching and research.The Roman Provincial collection has approximately 300,000 coins, and categorising these coins is a major task that requires many volunteer hours. Given that the tasks are manual and require the investment of multiple volunteer hours, GLAM sought a Machine Learning solution that would reduce the time required by a research department to identify and catalogue the coins as well as identify the basic attributes of the coin such as whether the image is on the reverse or obverse side of the coin. The model may also classify and categorise the coin in the dataset for similar searches.
Amazon Web Services offers a superior Platform that provides security and scale support through a broad architecture for multiple machine learning frameworks. With AWS services like Amazon Sagemaker a fully managed solution that offers cloud-based machine learning models to develop, train and deploy.
cloudmantra has a proven record in developing several highly scalable and efficient Machine Learning models for various use cases across Manufacturing, Academia, BFSI, Media industries. Oxford GLAM’s test bed project required quick response time for implementing the machine learning use cases presented with high accuracy and required high technical expertise.
The cloudmantra team first created the Roman Provincial Coinage dataset for model training using publicly available internet data on the RPC official website. The data was obtained using open-source web scraping technologies and stored in AWS S3. The data was then further categorized before being used to train the model using a similarity search model.
After constructing the dataset, the team utilised Amazon Sagemaker notebooks to construct and train the models. The solution provided a user interface through which one could upload an image of a coin, which was then pre-processed using object identification, auto-rotate, and upscale. Picture processing and prediction would follow, which would effectively identify the image for the features of the coin, such as the coin’s face, metal kind, area, and so on.
Using the Similarity search model, the algorithm was able to successfully predict the image and categorise it as a reference to a similar coin. Because this solution was developed for high availability and scalability, the GLAM team plans to combine it with a wider audience’s online portal in the future. The solution must be trained further using all of the datasets available on the Roman Provincial Coinage online dataset.
Volunteers often spend several hours analysing a coin, thus the machine learning model frees up many hours for them to dedicate to research and other important activities.
Machine Learning solution deployed within 6 weeks
Coin analysis completed in minutes that would take up to several hours
High Accuracy search for similar coins from the RPC database