Algorithm for Detecting Hidden Hazards of Crane Hook Safety Latch Based on Cascade Network

In order to carry out all-weather intelligent monitoring of whether the safety buckle status of crane hooks at construction sites is regulated, a cascading positioning and classification dual network detection algorithm combining traditional image optimisation and deep learning is proposed. Among them, the image optimization algorithm improves the detection ability of the model at night through adaptive cropping of the grey scale histogram values of the bright part of the image and dark area enhancement; the localisation module takes YOLOv7 as the main body, reduces the number of parameters of the model by replacing its backbone network with the lightweight EfficientNet B3, then introduces the CBAM attention mechanism to increase the localisation accuracy of the crane hook area. The loss function is replaced with SIoU that takes the angle factor into account to improve the convergence performance of the model; the classification module consists of the lightweight EfficientNet V2 network combined with the CA attention mechanism. In order to verify the reliability and superiority of the algorithm, a series of comparison experiments are conducted on a homemade construction site dataset. The results show that: the localisation model can reach 97.2% precision rate, with an inference time of 17.4ms, and also has good detection ability at night; the classification model achieves an overall precision and recall rate of 99.1% and 98.9%, with an inference time of only 8.1ms. It is thus evident that the overall structure of the lightweight cascaded dual network combined with the traditional image optimisation algorithm can achieve a good balance of detection precision and speed, while it can realise more efficient and accurate realtime online detection of hook safety buckle status in daytime, evening and nighttime conditions.