Enhanced Concealed Object Detection Method for MMW Security Images Based on YOLOv8 Framework With ESFF and HSAFF

Abstract

Millimeter-wave (MMW) imaging technology has been widely used in crowded areas such as airports and railway stations for concealed object detection (COD), owing to its characteristics of privacy and safety. However, the low signal-to-noise ratio (SNR) and low resolution in MMW security images lead to indistinct edges of concealed objects and a significant similarity to the human background. These limitations constrain the accurate and rapid detection of concealed objects on individuals. This article proposes a concealed objects detector that uses an enhanced You Only Look Once (YOLO) network, incorporating spatial, edge, and multiscale information to address the issues above. First, an efficient adaptive denoising method is designed to enhance image clarity. Second, considering the lack of prominent edge features in MMW images, an edge-spatial feature fusion (ESFF) module is introduced. This module enhances the network’s ability to learn edge features by combining them with spatial detail information. In addition, this article proposes a hierarchical scale-aware feature fusion (HSAFF) module to address the issue of high similarity between targets and background textures that impairs traditional detection networks, which can effectively reduce classification errors and false detections. Finally, the ESFF and HSAFF modules are integrated into the detection network based on the YOLOv8 framework. The experimental results on the MMW image dataset demonstrate that the proposed model effectively reduces classification and false detection losses, achieving mean average precision (mAP) @0.5 and mAP@ [0.5:0.95] of ${98}.{3}\%$ and ${81}.{5}\%$ , respectively, while the mAP of the proposed method is ${5}.{1}\%$ and ${4}\%$ higher than the baseline model, and surpassing other detection models.