Concealed Object Detection Based on Wavelet Convolution for Millimeter Wave Image

The detection of concealed objects in millimeter- wave (MMW) images has garnered considerable attention due to its non-contact and non-invasive nature, setting it apart from traditional security inspection methods. Nonetheless, the low image resolution inherent in this technology poses challenges in detecting small targets effectively. This research endeavors to address these limitations in the context of detecting safety- hazard objects carried by miners, who wear thick overalls, thus presenting a more complex detection scenario. To surmount these challenges, we propose a novel enhancement technique that leverages wavelet transform to augment existing detectors. The introduced approach, termed Wavelet-Conv, is integrated into the YOLOv8 architecture for enhanced detection performance. To ensure comprehensive evaluation, we meticulously curated a dedicated dataset with diverse instances of concealed objects and conducted extensive experiments to evaluate the proposed model. Results substantiate the efficacy of our method, as it significantly improves recall and mean average precision (mAP) in concealed object detection within MMW images, surpassing the performance of existing state-of-the-art methods.