Enhancing concealed object detection in active THz security images with adaptation-YOLO.

Area of Science:

• Security technology
• Image processing
• Artificial intelligence

Background:

• Terahertz (THz) security scanners are crucial for non-contact inspection and detecting dangerous goods, aiding in counter-terrorism efforts.
• Current object detection algorithms struggle with THz images due to small object sizes, low resolution, and background noise, often neglecting contextual object dependencies.

Purpose of the Study:

• To develop an accurate and efficient method for detecting concealed objects in THz security images.
• To address the limitations of existing object detection algorithms in handling the unique challenges of THz imagery.

Main Methods:

• Proposed an adaptive context-aware attention network (ACAN) to model global contextual features in spatial and channel dimensions, fusing local and global information.
• Developed a dynamic adaptive convolution block (DACB) to adaptively adjust convolution filters and suppress interference.
• Integrated ACAN and DACB into YOLOv8, creating the Adaptation-YOLO model.

Main Results:

• Adaptation-YOLO demonstrated significant improvements in detecting concealed objects within THz security images.
• The method effectively enhanced feature capture by modeling contextual dependencies and suppressing noise.
• Experimental results on an active THz image dataset confirmed the increased accuracy and efficiency of the proposed approach.

Conclusions:

• The developed Adaptation-YOLO model, integrating ACAN and DACB, offers a robust solution for concealed object detection in THz security imaging.
• This advancement holds promise for improving the effectiveness of security screening systems.
• The study highlights the importance of context-aware attention and adaptive convolution for challenging image analysis tasks.