徐州工程学院学报（自然科学版）2026，Vol.41Issue(1)：88-94,7.
基于深度卷积的建筑施工现场多目标危险行为识别方法
A Method for Identifying Multi-Objective Hazardous Behaviors in Construction Sites Based on Deep Convolution
摘要
Abstract
A multi-objective method for recognizing dangerous behaviors based on deep convolution is proposed to address errors in manually monitoring and identifying such behaviors on construction sites.The on-site multi-target dangerous behavior recognition problem is decomposed into two steps:on-site monitoring image encoding,and multi-target dangerous behavior decoding and recognition.During the encoding stage,dense convolution and cross-layer fusion techniques are employed to extract and combine multi-target behavioral features from construction site monitoring images.These features are then decoded,after which a Softmax classifier is used to determine the probability that the decoded behavioral features belong to a given behavior type,thus achieving multi-target dangerous behavior recognition.Experimental results demonstrate that the proposed method can accurately identify the subtle behavioral differences of multiple targets on construction sites.关键词
深度卷积神经网络/建筑施工现场/多目标危险行为识别/Softmax 分类/特征提取Key words
deep convolutional neural network/construction site/multi-objective identification of dangerous behaviors/Softmax classification/feature extraction分类
信息技术与安全科学引用本文复制引用
吴芳芳..基于深度卷积的建筑施工现场多目标危险行为识别方法[J].徐州工程学院学报（自然科学版）,2026,41(1):88-94,7.基金项目
2023年安徽省高校科研重点项目(2023AH053119) （2023AH053119）
