Unsafe hoisting behavior recognition for tower crane based on transfer learning

Tower cranes commonly encounter safety accidents related to unsafe hoisting behaviors on construction sites globally. Effectively monitoring unsafe hoisting behaviors has become a challenging aspect in the safety management of tower cranes. Consequently, this paper introduces a recognition framework based on transfer learning to identify unsafe hoisting behaviors of tower cranes, specifically tilt hoisting, sudden braking, and sudden unloading. The model architecture is developed through deep adversarial domain adaptation. Experimental results demonstrate that the proposed transfer learning model achieves a recognition accuracy of 76.74%, outperforming other methods. It effectively mitigates the negative transfer phenomenon arising from the absence of a target domain sample dataset. This research is of practical significance in enhancing safety management practices related to tower crane hoisting on construction sites. In the future, the model can be extended to various hoisting conditions to accumulate domain knowledge.