Latest Deep learning-Based Research Provides More Rapid And Accurate Terahertz Security Inspections Using Human Image Data

With the worldwide anti-terrorist measures being strengthened, it is becoming more vital to undertake security checks in public locations to discover concealed items on the human body. Past research has shown that deep learning may assist in identifying hidden items in passive terahertz images. However, real-time tagging with high accuracy and performance remains a challenge. 

In a publication in Scientific Reports, Prof. Fang Guangyou and his research team from the Aerospace Information Research Institute and Chinese Academy of Sciences used human image data obtained by passive terahertz sensors, and they trained and tested a potential detector based on deep residual networks. The suggested approach can be utilized to identify concealed items in terahertz pictures in real-time.

To lessen the complexity of network training, the research group swapped the backbone network of the Single Shot MultiBox Detector (SSD) method with a more representative residual network. A feature fusion-based terahertz image target identification method was presented to address the issues of repetitive detection and missed detection of tiny objects, with an addition of a hybrid attention mechanism to SSD to boost the algorithm’s ability to collect object details and position information.

The research group also compared the suggested model to other standard detection approaches on the terahertz human security picture dataset. When the speed was slightly lowered, the findings revealed that the proposed technique achieves better detection accuracy than the original SSD algorithm.

The enhanced SSD method solves missed detection while simultaneously increasing detection confidence. As a result, it can meet the real-time detection requirements of security inspection situations.

This Article is written as a research summary article by Marktechpost Staff based on the research paper 'Improved SSD network for fast
concealed object detection and recognition in passive terahertz security images'. All Credit For This Research Goes To Researchers on This Project. Checkout the paper and reference article.

Please Don't Forget To Join Our ML Subreddit