With the increasing complexity of the global security landscape, vehicle security inspection at border ports faces severe challenges. Traditional under-vehicle hazardous material detection relies on manual inspection or basic image processing techniques, which suffer from low efficiency, low accuracy, and poor environmental adaptability. This paper proposes a CBAM-YOLOv5 object detection algorithm that combines the Convolutional Block Attention Module (CBAM) with the YOLOv5 network. This method introduces depthwise separable convolution (DWConv) and convolutional block attention module from the YOLOv5 algorithm framework, and optimizes the loss function the detection network from three aspects to improve YOLOv5. Research demonstrates that, compared to the original model, the improved YOLOv5 network model achieves enhancements in both precision and mean Average Precision (mAP). Specifically, it attains a mean Average Precision (mAP) of 95.2% on a self-built under-vehicle border security dataset. This study provides technical support for improving the efficiency of border vehicle security inspection and advancing the intelligence of border control.