|国家预印本平台
首页|基于增强语义信息分析的室内场景识别方法

基于增强语义信息分析的室内场景识别方法

Enhanced Semantic Information Extraction for Indoor Scene Recognition

中文摘要英文摘要

针对室内场景识别任务面临的复杂性挑战,提出了一种基于增强语义信息分析的场景识别方法。结合语义分割算法从空间维度建模语义标签及其相互依赖性,设计了一种语义歧义过滤策略,通过自适应判别域的置信度过滤机制提升语义标签的可靠性,通过通道维度分组的并行子网络进行语义的特征提取。方法还利用特征融合方法将将语义特征与色彩特征进行多尺度整合,并通过注意力机制增强模型对场景中重要信息的表达能力。实验结果表明,在ADE20k和MIT Indoor两个数据集上,本文提出的基于增强语义信息分析的室内场景识别方法的准确率分别为57.40%和88.93%,优于大部分现有的场景识别算法,为室内场景识别提供有效支持。

In this paper,to address the complexity challenges of indoor scene recognition tasks, an enhanced semantic information analysis-based scene recognition method is proposed. The method leverages semantic segmentation algorithms to model semantic labels and their interdependencies from a spatial perspective. A semantic ambiguity filtering strategy is designed, employing a confidence filtering mechanism based on adaptive decision domains to improve the reliability of semantic labels. Semantic feature extraction is performed using parallel sub-networks grouped by channel dimensions. Additionally, a feature fusion approach is utilized to integrate semantic and color features across multiple scales, and attention mechanisms are adopted to enhance the model\'s representation of critical scene information. Experimental results demonstrate that the proposed enhanced semantic information analysis-based indoor scene recognition method achieves accuracy rates of 57.40% and 88.93% on the ADE20k and MIT Indoor datasets, respectively, outperforming most existing scene recognition algorithms and providing effective support for indoor scene recognition.

王云帆、伍淳华

计算技术、计算机技术

人工智能场景识别语义关系深度学习计算机视觉

artificial intelligencescene recognitionsemantic relationdeep learningcomputer vision

王云帆,伍淳华.基于增强语义信息分析的室内场景识别方法[EB/OL].(2025-02-13)[2025-08-18].http://www.paper.edu.cn/releasepaper/content/202502-15.点此复制

评论