Scholarlink

<p>The interpretation of cultural artifacts using AI is gaining significant attention for heritage preservation and analysis. However, existing deep learning model with poor generalization high computational complexity and limited interpretability as they are applied to diverse artifacts datasets. To overcome these challenge DeepCaps-Transformer integrates capsule network and transformer models to provide an efficient accurate and explainable framework for cultural artifact understanding. The proposed DeepCaps-Transformer framework performs purely visual artifact classification and does not include any generative or GPT-style text interpretation components. Initially, the input images are collected from the Cultural Heritage Visual dataset. The input images are preprocessed utilizing Synthetic Minority Over-sampling Technique (SMOTE) and Adaptive Synthetic Sampling (ADASYN) to balance class distributions and enhance model generalization. Then, visual feature extraction is performed using Squash convolutional neural network (CNN) a lightweight convolutional architecture optimized for capturing key stylistic patterns and artifact contours with fewer parameters. For classification, a hybrid model combining Deep Capsule Network (DCN) and Transformer Encoder with Channel-wise Attention is utilized. The Capsule Network captures spatial hierarchies and orientation features, although the Transformer integrates Squeeze-and-Excitation (SE) module to enhance salient channel information. Finally, the loss function is optimized using novel Hybrid Secretary Bird with Fossa Optimization Algorithm (HSBFOA).The techniques are implemented on the Python platform and performances are compared with various existing methods based on high accuracy (99.12%) and low misclassification rate (1.40%). Overall, the proposed DCN demonstrates effectiveness in providing accurate, reliable and context aware interpretation of cultural artifacts highlighting its potential for robust cross-domain artifact analysis.</p>

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

DeepCaps-Transformer: Cultural Artifact Interpretation Using Deep Capsule Networks and Transformer Models

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Vanishree Kavi Mahesh,
  • Jambukeshwar Pujari

摘要

The interpretation of cultural artifacts using AI is gaining significant attention for heritage preservation and analysis. However, existing deep learning model with poor generalization high computational complexity and limited interpretability as they are applied to diverse artifacts datasets. To overcome these challenge DeepCaps-Transformer integrates capsule network and transformer models to provide an efficient accurate and explainable framework for cultural artifact understanding. The proposed DeepCaps-Transformer framework performs purely visual artifact classification and does not include any generative or GPT-style text interpretation components. Initially, the input images are collected from the Cultural Heritage Visual dataset. The input images are preprocessed utilizing Synthetic Minority Over-sampling Technique (SMOTE) and Adaptive Synthetic Sampling (ADASYN) to balance class distributions and enhance model generalization. Then, visual feature extraction is performed using Squash convolutional neural network (CNN) a lightweight convolutional architecture optimized for capturing key stylistic patterns and artifact contours with fewer parameters. For classification, a hybrid model combining Deep Capsule Network (DCN) and Transformer Encoder with Channel-wise Attention is utilized. The Capsule Network captures spatial hierarchies and orientation features, although the Transformer integrates Squeeze-and-Excitation (SE) module to enhance salient channel information. Finally, the loss function is optimized using novel Hybrid Secretary Bird with Fossa Optimization Algorithm (HSBFOA).The techniques are implemented on the Python platform and performances are compared with various existing methods based on high accuracy (99.12%) and low misclassification rate (1.40%). Overall, the proposed DCN demonstrates effectiveness in providing accurate, reliable and context aware interpretation of cultural artifacts highlighting its potential for robust cross-domain artifact analysis.