Abstract

In the construction industry, about 80–90% of accidents are caused by the unsafe actions and behaviors of employees. Thus, behavior management plays a key role in enhancing safety. In particular, behavior observation is the most critical element for modifying workers’ behavior in a safe manner. However, there is a lack of practical methods to measure workers’ behavior in construction as current literature only focuses on a few unusual signs such as not wearing personal protective equipment. This paper proposes a system for recognizing workers’ dangerous behaviors. To that end, an image dataset has been collected, labeled for three such behaviors. Based on the dataset obtained, the transfer-learning approach is used with three pre-trained models, VGG19, Inception_V3 and InceptionResnet_V2. The results indicate that InceptionResnet_V2 performs better than VGG19_ and Inception_V3 for classifying unsafe behaviors and after 150 epochs, its accuracy reaches 92.44%.

中文翻译：

---

使用深度卷积神经网络进行不安全施工行为分类

摘要

在建筑行业，大约 80-90% 的事故是由员工的不安全行为和行为引起的。因此，行为管理在提高安全性方面起着关键作用。特别是，行为观察是以安全的方式改变工人行为的最关键因素。然而，由于目前的文献仅关注少数不寻常的迹象，例如未佩戴个人防护设备，因此缺乏衡量建筑工人行为的实用方法。本文提出了一种识别工人危险行为的系统。为此，我们收集了一个图像数据集，标记了三种这样的行为。基于获得的数据集，迁移学习方法与三个预训练模型一起使用，VGG19、Inception_V3 和 InceptionResnet_V2。