基于深度学习的轻量级自然语言描述生成算法  

Lightweight Natural Language Description Generation Algorithm Based on Deep Learning

作　　者：李健[1] 马延周 胡瑞娟 LI Jian;MA Yan-zhou;HU Rui-juan(PLA Strategic Support Force Information Engineering University,Luoyang Henan 471003,China)

机构地区：[1]战略支援部队信息工程大学,河南洛阳471003

出　　处：《计算机仿真》2022年第10期369-372,390,共5页Computer Simulation

基　　金：国家自然科学基金重大项目(11590771)。

摘　　要：为了提升轻量级自然语言描述生成效果与效率,以视频图像作为研究对象,提出基于深度学习的轻量级自然语言描述生成算法,分析自然语言生成后能否对视频图像内容作出充分解读。采用深度迁移学习算法提取视频图像中的语义特征与帧流语义特征,建立多特征融合的轻量级自然语言描述生成模型,对所提取的图像语义特征与帧流语义特征实施融合后,导入视频自然语言描述模型,生成自然语言描述。实验结果表明,所提算法不受视频图像类型、数量的约束,在不同条件下所生成的视频图像轻量级自然语言描述均符合视频图像内容,且对视频图像特征的提取、融合效果较好,可为轻量级自然语言描述生成工作提供有效协助。This paper proposes a lightweight natural language description generation algorithm based on deep learning for improving the effect and efficiency of lightweight natural language description generation. After the generation of natural language, the interpretation of video image content was analyzed in detail. Deep transfer learning algorithm was applied to extract semantic features of video image and frame stream. The multi-feature fusion lightweight natural language description was founded to generate the model. The extracted image semantic features and frame stream semantic features were combined to import into the video natural language description model to generate natural language description. The results show that the algorithm is not facile to be affected by the type and quantity of video images, and the generated video image lightweight natural language description is consistent with the actual video image content and has excellent extraction and fusion effect, implying that it has a good application prospect in lightweight natural language description generation.

关 键 词：深度学习 轻量级 自然语言 描述生成 特征提取 特征融合 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]