BDSLI：一种用于孟加拉手语翻译的混合CNN-Transformer模型
BDSLI: A hybrid CNN-Transformer model for Bengali Sign Language interpretation
- BRAC University（BRAC大学）
机构由 AI 辅助整理，请以论文原文为准。
中文总结 AI 辅助
针对孟加拉手语识别进展有限的问题，提出混合CNN-Transformer模型，在自定义62词数据集上达到98.65%测试准确率，并部署于网络应用验证。
中文摘要 AI 辅助
本研究引入了一种新颖的混合CNN-Transformer架构，以解决孟加拉手语识别（SLR）领域进展有限的问题，重点关注孤立手语词识别和句子生成。这种特定的模型组合在孟加拉手语识别任务中是全新的。我们开发了一个自定义视频数据集，包含62个不同的孟加拉手语词（每类250个样本），以及一个独立的测试数据集。该CNN-Transformer模型在所有对比模型和基线模型（如CNN-LSTM、独立TCN）中表现出优越的性能，达到了99.58%的训练准确率（验证准确率99.48%）和98.65%的测试准确率。训练后的模型随后部署在一个网络应用程序中，用于实际场景验证。
英文摘要
This study introduces a novel hybrid CNN-Transformer architecture to address the limited progress in Bengali SLR, focusing on isolated sign word recognition and sentence generation. This specific model combination is new to Bengali SLR tasks. A custom video dataset was developed, featuring 62 distinct Bengali sign words (250 samples/class), along with a separate test dataset. The CNN-Transformer model demonstrated superior performance against all comparative and baseline models (e.g., CNN-LSTM, standalone TCN), achieving a 99.58% training accuracy (99.48% validation) and a 98.65% test accuracy. The trained model was subsequently deployed in a web application for real-world validation.
