Scholarlink

Studying ancient languages and translating them is a trending area of research in the realm of research, one of the most prominent fields is the study and translation of ancient languages. The process of translating cuneiform tablets and interpreting them with the use of various artificial intelligence algorithms has been the subject of a significant amount of research. Nevertheless, one of the challenges that continues to be encountered is the quantity of data that is required for training techniques that involve deep learning algorithms. Therefore, in every research project, efforts are made to collect as much information as possible to achieve the highest possible level of precision in the findings. First, work is done to collect image data for the cuneiform tablets (both Assyrian and Sumerian) to construct a database for classification between Sumerian and Assyrian. Secondly, work is done to create a detection of 11 signs of the New Assyrian writing and the pronunciation of each sign. Both of these tasks are part of this research. The research utilized YOLOv8, a pre-trained model commonly employed in computer vision applications and recognized as a contemporary approach. It was utilized to categorize and identify ancient cuneiform inscriptions from which contemporary cuneiform scripts were derived. Approximately, 900 photographs of Assyrian tablets were gathered from the Museum in Iraq and increased to almost 2000 photographs by utilizing preprocessing and enhancing techniques. We developed a unique dataset and subsequently conducted training on the model using it. Also, a collection of images depicting different cuneiform (Neo-Assyrian, Babylonian, and Sumerian) scripts. It was gathered from ancient books and an Iraqi museum. A classification was conducted between Assyrian and other cuneiform scripts, and a sample of neo-Assyrian scripts was detected. Each script was examined by training the algorithm and noting the pronunciation of each letter and sign. In this paper, the mean average percentage (mAP) was used to measure the accuracy performance of the detection model, which reached 82%; in the classification model, the accuracy reached 96%.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Detection and Classification of Ancient Cuneiform Script Writing Using the YOLOv8 Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Elaf A. Saeed,
  • Ammar D. Jasim,
  • Munther A. Abdul Malik

摘要

Studying ancient languages and translating them is a trending area of research in the realm of research, one of the most prominent fields is the study and translation of ancient languages. The process of translating cuneiform tablets and interpreting them with the use of various artificial intelligence algorithms has been the subject of a significant amount of research. Nevertheless, one of the challenges that continues to be encountered is the quantity of data that is required for training techniques that involve deep learning algorithms. Therefore, in every research project, efforts are made to collect as much information as possible to achieve the highest possible level of precision in the findings. First, work is done to collect image data for the cuneiform tablets (both Assyrian and Sumerian) to construct a database for classification between Sumerian and Assyrian. Secondly, work is done to create a detection of 11 signs of the New Assyrian writing and the pronunciation of each sign. Both of these tasks are part of this research. The research utilized YOLOv8, a pre-trained model commonly employed in computer vision applications and recognized as a contemporary approach. It was utilized to categorize and identify ancient cuneiform inscriptions from which contemporary cuneiform scripts were derived. Approximately, 900 photographs of Assyrian tablets were gathered from the Museum in Iraq and increased to almost 2000 photographs by utilizing preprocessing and enhancing techniques. We developed a unique dataset and subsequently conducted training on the model using it. Also, a collection of images depicting different cuneiform (Neo-Assyrian, Babylonian, and Sumerian) scripts. It was gathered from ancient books and an Iraqi museum. A classification was conducted between Assyrian and other cuneiform scripts, and a sample of neo-Assyrian scripts was detected. Each script was examined by training the algorithm and noting the pronunciation of each letter and sign. In this paper, the mean average percentage (mAP) was used to measure the accuracy performance of the detection model, which reached 82%; in the classification model, the accuracy reached 96%.