Multi-Object Detection in Security Screening Scene Based on Convolutional Neural Network.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Convolutional neural networks (CNNs) are widely used for target detection.
• Current CNNs for X-ray security screening lack sufficient accuracy.
• Existing methods struggle with feature extraction and redundant information.

Purpose of the Study:

• To improve the accuracy of target detection in X-ray images for security screening.
• To introduce a novel architecture for enhanced feature extraction and attention.
• To integrate the proposed architecture into the Single Shot MultiBox Detector (SSD) algorithm.

Main Methods:

• Implemented ResNet as a backbone network to replace VGG for superior feature extraction.
• Designed a multi-scale feature extraction (MSE) structure to enrich feature layer information.
• Integrated a multi-scale attention architecture (MSA) to refine features and contextual information.
• Utilized a combination of Adaptive-Non-Maximum Suppression (NMS) and Soft-NMS for final bounding box prediction.

Main Results:

• The proposed coupled multi-scale architecture significantly improved target detection effectiveness.
• The mean average precision (mAP) value increased by 7.4% compared to the original SSD approach.
• The enhanced detection accuracy was achieved while maintaining the original detection speed.

Conclusions:

• The novel architecture effectively addresses limitations in current X-ray target detection systems.
• The integration of MSE and MSA modules enhances feature representation and reduces interference.
• This approach offers a more accurate and efficient solution for security screening applications.