LAB-CNN: LoD-specific Attention-based Branch Convolutional Neural Network for Digital Twin

Ji-Wan Kim, Kyu-Sik Kim, Hyun-Suk Lee

Abstract: In this paper, we address a digital twin (DT) synchronization, which is necessary for DT tasks, considering the level of detail (LoD) using hierarchical details about real-world objects. To this end, we propose LoD-specific Attention-Based Convolutional Neural Network architecture (LAB-CNN) to address the LoD-aware classification problem for DT synchronization. The LAB-CNN architecture is composed of a traditional CNN model with branch networks for each LoD. This can output an attention map and predictions for each level, considering each LoD's objectives. Through simulation, we demonstrate that our LAB-CNN not only detects suitable attention locations for each LoD's objectives but also performs high-accuracy prediction across all LoDs compared to the traditional hierarchical classification model.

External IDs:dblp:conf/metacom/KimKL24