基于深度学习的施工安全泛场景数据获取方法

发布时间：2018-04-09 04:41

本文选题：深度学习　切入点：施工场景　出处：《中国安全科学学报》2017年05期

【摘要】：为实现隐患场景识别和施工安全泛场景数据智能化处理,在分析场景构成因素基础上,将施工场景划分为不安全行为场景和不安全物态场景2类。结合图像语义层次的划分,应用深度学习方法提取对象语义、空间关系语义、场景语义和行为语义,进而依据施工安全泛场景数据理论,得到图像语义信息与泛场景数据对应关系。选取建筑施工现场照片为样本源,采集施工不安全行为场景数据,验证应用基于深度学习的方法处理泛场景数据的可行性。结果表明:通过深度学习提取图像语义,可以自动采集不安全行为和不安全场景泛物态数据,提高场景数据处理效率。
[Abstract]:In order to realize the identification of hidden trouble scene and the intelligent processing of the universal scene data of construction safety, the construction scene is divided into two categories: unsafe behavior scene and unsafe material state scene, based on the analysis of the factors of scene composition.Combined with the classification of image semantic level, the object semantics, spatial relation semantics, scene semantics and behavior semantics are extracted by using the depth learning method, and then based on the construction safety pan-scene data theory.The relationship between image semantic information and pan-scene data is obtained.The construction site photos are selected as the sample source to collect the construction unsafe behavior scene data and verify the feasibility of using the method based on depth learning to deal with the pan-scene data.The results show that extracting image semantics through deep learning can automatically collect unsafe behavior and pan-state data of unsafe scenes and improve the efficiency of scene data processing.
【作者单位】：中国矿业大学(北京)资源与安全工程学院;
【基金】：国家自然科学基金资助(51674268)
【分类号】：TP391.41;TU714

【参考文献】