A Complex Scenario-Oriented Intelligent Monitoring Method for Road Geohazards with YOLOv5-CBCF

Geological disasters such as avalanches and landslides around road routes pose a great threat to transportation. To address the problems of high complexity, high false alarm rate, and low timeliness of the traditional landslide collapse monitoring method, a road rockfall and landslide monitoring model method based on the improved YOLOv5-CBCF is proposed, which introduces the lightweight downsampling method ContextGuided by improving and replacing the Conv downsampling module in the Head layer to increase the sensing field of the feature map, and also increase the model's The sensing range of the feature map is increased, the detection speed is improved, a novel two-layer routing mechanism (BiFormer) is introduced in the Head side to improve the model monitoring accuracy, and the Focaler-IoU function is introduced to optimize the model regression performance by focusing on the regression samples with different difficulty levels and dynamically adjusting the weights of the samples. The accuracy of YOLOv5-CBCF for road landslide detection is improved from 79.27 % to 83.73 %, the Map50 value is improved from 78.86 % to 82.48 %, and the recall rate is improved from 78.1 % to 82.7 %. The accuracy of road rockfall detection was improved from 79.74 % to 84.42 %, the Map50 value was improved from 77.28 % to 82.75 %, and the recall rate was improved from 78.2 % to 82.3 %. The results show that the improved model effectively enhances the comprehensive identification detection effect