Underwater Detection of Ancient Pottery Sherds Using Deep Learning

Konstantinos Paraskevas, Ioannis Mariolis, Georgios Giouvanis, Georgia Peleka, Georgios Zampokas, Dimitrios Tzovaras, Centre for Research & Technology Hellas, Information Technologies Institute, Thessaloniki, Greece

ABSTRACT

This paper presents the development of a machine learning model for the detection of ancient pottery shreds in the vicinity of an underwater shipwreck in Modi Island, Greece. Specifically, multiple versions of the YOLOv8 model were trained using a custom image dataset created from videos captured at the shipwreck site during diving expeditions. The objective of the presented research is to deploy the developed object detection system in a remotely operated vehicle (ROV) to facilitate the automatic identification of pottery shreds and support archaeological excavations. The paper details the methodology employed during the model development process and presents the results achieved through extensive experimentation and evaluation. The findings highlight the ef ectiveness and potential of the proposed model in enhancing the ef iciency and accuracy of archaeological exploration and analysis in underwater environments.

KEYWORDS

Ancient pottery shreds detection, underwater archaeological excavations, machine learning, object detection, remotely operated vehicle (ROV), underwater shipwrecks, YOLOv8 model.

An Improved Teaching-learning-based Optimization With Adaptive Moment Estimation for Global Optimization

XiaoKui Wu¹ and ZhuRong Zhou², ¹SOUTHWEST UNIVERSITY COLLEGE OF COMPUTER & INFORMATION SCIENCE Chongqing, CHINA, ²COLLEGE OF COMPUTER & INFORMATION SCIENCE Chongqing, CHINA

ABSTRACT

An Improved Teaching-Learning-Based Optimization with adaptive moment estimation (TLBO-Adam) to solve the problem that the Teaching-Learning-Based Optimization is easy to appear premature convergence and slow convergence speed. Based on the adaptive moment estimation which has the characteristics of goal-oriented mechanism, the convergence process of the algorithm is accelerated. At the same time, the fuzzy c-means algorithm (FCM) is used to group the initial population, and different teachers are assigned to each group to avoid all students learning from one teacher. In this way, the diversity of the population is maintained and the premature convergence of the algorithm is avoided. Experimental optimization problem for 7 single peak and 2 typical multimodal optimization problem is optimized, and compared with other swarm intelligence algorithms, the experimental results show that the improved algorithm TLBO-Adam in the 9 benchmark functions, can quickly and efficiently find the global optimal solution, is better than that of the Teaching-Learning-Based Optimization algorithm.

KEYWORDS

Teaching–learning-based optimization, Fuzzy c-means algorithm, Adaptive Moment Estimation, Globally Optimal Solution.