Lightweight prohibited item detection method based on YOLOV4 for x-ray security inspection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Security Technology

Background:

• Deep learning methods have shown promise in detecting prohibited items via X-ray security inspection.
• Existing deep learning object detection models often have high parameter counts and computational demands, limiting practical application in security settings.

Purpose of the Study:

• To propose a lightweight deep learning model for efficient prohibited item detection in X-ray security inspection.
• To reduce the computational complexity and hardware requirements of object detection methods for X-ray security applications.

Main Methods:

• Developed a lightweight prohibited item detection method using YOLOV4 as a base.
• Replaced the YOLOV4 backbone with MobilenetV3 and optimized the neck and head using depthwise separable convolutions.
• Integrated an adaptive spatial-and-channel attention block to enhance feature extraction.
• Utilized focal loss to address class imbalance during training.

Main Results:

• The proposed method achieved a mean average precision 4.98% higher than YOLOV4-tiny and 0.07% lower than YOLOV4 on a real X-ray pseudocolor image dataset.
• The number of parameters and computational consumption were significantly reduced compared to YOLOV4, while remaining slightly higher than YOLOV4-tiny.

Conclusions:

• The lightweight YOLOV4-based method offers a practical solution for prohibited item detection in X-ray security inspections.
• The model balances detection accuracy with reduced computational cost, making it suitable for resource-constrained security hardware.