Abstract: In the process of open-pit mine transportation,phenomena such as light vehicles running without tickets and
human-caused ticket fraud occur frequently,which significantly reduces the authenticity and reliability of the transportation data
statistics,and is not conducive to the operation and management of the mine. By applying image recognition technology,a method
for detecting the loading rate of open-pit mine trucks based on the improved YOLOv5s is proposed. The open-pit mine truck
loading image dataset is enhanced and expanded,and then labeled. On the basis of the YOLOv5s network structure,the improved
backbone network GhostNet is used for feature extraction. The shallow network P2 is added to refine the feature output,
enhancing the network′s ability to effectively capture spatial information. At the same time,the throughput configurable convolution
C2f module is introduced to ensure lightweight while obtaining richer gradient flow information. In the post-processing
stage of object detection,the smoother soft-NMS algorithm is used to replace the NMS algorithm to remove redundant detection
boxes,and the loss function CIoUα is used to calculate the loss of the rectangular box. The research results show that the improved
YOLOv5s model has recognition accuracies of 83. 2%,90. 4%,93. 3%,92. 4%,and 94. 1% for trucks with different
loading rates (70%,80%,90%,100%,and 110%),which can meet the on-site monitoring needs of the mine. This method has
the characteristics of non-contact measurement objects,no interference with the transportation system,low operating costs,and
no need for manual supervision. This method is characterized by non-contact measurement,non-interference with the transportation
system,low operating costs,and no need for manual supervision,which can provide data support for the refined management
of open-pit mine transportation.