Risk detection in building construction scenarios based on ResNet-LSTM

Construction is characterized by many types of work, intensive use of electrical equipment, and complex high-altitude operations. Effective risk detection of construction scenarios is the key to ensuring personnel safety and construction quality. In this regard, this study proposes an intelligent construction scene risk detection technology. This study proposes a welding scene risk detection technology based on ResNet-LSTM. By improving the ResNet feature module, a spatial attention layer is introduced to enhance key feature capture. It proposes a high-altitude scene risk detection technology based on an improved ResNet and GRU, uses ResNet that integrates dual attention mechanisms to enhance detailed feature extraction, and uses GRU to enhance character temporal feature extraction to improve the capture effect of high-altitude features. In the experimental analysis in the welding scene, the accuracy of the research model reaches 93.5% and the average precision is 93.2%, which is better than similar models. In high-altitude scene detection, the research model’s detection accuracy in the safety rope-fixed scene with 20% occlusion rate is 82.3%, which is better than similar models. In multi-scenario risk detection, the research model is closest to the measured value in the risk assessment of 8 risk scenarios, with a maximum deviation of 0.34, and has the best performance. The research technology has good application effects in the safety detection of building construction scenes. This study provides technical support for building safety management and construction quality supervision.