02918nas a2200361   4500000000100000008004100001653001900042653001700061653003300078653003300111653003400144653001800178653001800196653002500214653002000239653001500259653001400274653002300288653002900311653003000340653002100370653002200391653001600413653001600429653001300445653002200458100001300480700001500493245011700508856015600625300001200781520176300793                                       d10aClassification10aClient sides10aconvolutional neural network10aconvolutional neural network10aConvolutional neural networks10aDeep learning10aDeep learning10aDigital preservation10aDigital storage10aE-learning10aF1 scores10aFederated learning10aIntangible heritage coin10aIntangible heritage coins10aLearning systems10aModeling approach10aPerformance10aSide models10aTextures10aTransfer learning1 aS. Mehta1 aV. Kukreja00aHeritage Coins Classification: An Federated CNN Approach to Analyse Performance of Global and Client-side Models  uhttps://www.scopus.com/inward/record.uri?eid=2-s2.0-85190533164&doi=10.1109%2fICSH57060.2023.10482817&partnerID=40&md5=9dd1924093c5875a3c4918b35158ec2f  a141-1463 aUsing federated learning and CNN-based approaches, this research suggests a unique digital preservation and categorization method of intangible heritage coins. We compare the performance of a global model approach to a client-side model method and discover that the worldwide model approach outperforms the client-side model approach in terms of accuracy, precision, recall, and F1 score. In particular, the client-side model method obtains an accuracy of 89.8\%, a precision of 88.3\%, a remembrance of 90.5\%, and an F1 score of 89.3\%, compared to the global model approach s accuracy, precision, recall, and F1 score of 92.7\%, 91.5\%, and 92.5\%. Our method provides a viable option for digitally preserving intangible heritage coins, allowing us to protect private information and preserve these priceless artefacts for future generations. Our findings demonstrate that our method successfully categorizes ethereal heritage coins, with the most challenging classes to categories being those with identical characteristics or patterns. This study contributes to organising and protecting cultural assets by outlining a novel strategy that uses federated learning and deep learning approaches. We give a thorough description of the methods we utilized for our study, including data collection, model training, and assessment, in addition to the findings of our performance evaluation. Our study advances the area of heritage categorisation and preservation by presenting a novel strategy that uses federated learning and deep learning technologies. Future work may study other deep learning models, examine how well our method scales to more enormous datasets, and enhance the categorisation of difficult classes with minor texture or shading variations.