HoloMine: A Synthetic Dataset for Buried Landmines Recognition using Microwave Holographic Imaging

TL;DR

The paper addresses the risk-intensive problem of buried landmine detection by introducing HoloMine, a large synthetic dataset of microwave holographic images for recognition tasks. It builds 2D holograms and 3D inverted maps by fusing indoor and outdoor scans of six mine replicas, clutter, and pottery, totaling 41,800 samples, with annotations for classification, detection, and segmentation. It leverages $H = \alpha H^{IN} + (1-\alpha) H^{OUT}$ with $\alpha^* \approx 0.14$, and uses angular-spectrum reconstruction to produce the 3D holograms, providing a publicly available benchmark. The experiments show that 2D holographic representations generally outperform 3D reconstructions, revealing room for methodological improvements and the value of multi-sensor extensions. Overall, HoloMine offers a scalable, openly accessible resource to advance automated demining research and reduce operational risk.

Abstract

The detection and removal of landmines is a complex and risky task that requires advanced remote sensing techniques to reduce the risk for the professionals involved in this task. In this paper, we propose a novel synthetic dataset for buried landmine detection to provide researchers with a valuable resource to observe, measure, locate, and address issues in landmine detection. The dataset consists of 41,800 microwave holographic images (2D) and their holographic inverted scans (3D) of different types of buried objects, including landmines, clutter, and pottery objects, and is collected by means of a microwave holography sensor. We evaluate the performance of several state-of-the-art deep learning models trained on our synthetic dataset for various classification tasks. While the results do not yield yet high performances, showing the difficulty of the proposed task, we believe that our dataset has significant potential to drive progress in the field of landmine detection thanks to the accuracy and resolution obtainable using holographic radars. To the best of our knowledge, our dataset is the first of its kind and will help drive further research on computer vision methods to automatize mine detection, with the overall goal of reducing the risks and the costs of the demining process.

Figures (8)

• Figure 1: Detail of a PMN-4 scan represented in 2D complex holographic image $H^0$ and 3D reconstruction $\overline{H}$. On the right, is presented amplitude and phase of $H^z = f(H^0, z)$.
• Figure 2: left) view of the 3-D mechanical arm and Holographic Subsurface Radar (HSR); right) a panoramic view of the Robotic platform. This system has been used to acquire the basic elements used to create the dataset.
• Figure 3: Transformation from zig-zag time-serie acquisition $S,P$ to $H$ through $T:S,P \rightarrow H$.
• Figure 4: Dataset creation pipeline from indoor and outdoor scan acquisition [left] to pre-processing [middle] to 2D and 3D image creation [right]. $w$ indicates the weight used to fuse the hologram images.
• Figure 5: Objects considered for indoor scans. On the first row, landmines replicas are reported, namely (from left to right): PMN-4(d=95mm, h=46mm), PMN-1(d=95mm, h=55mm), VS-50(d=90mm, h=45mm), TYPE 72(d=78mm, h=38mm), M-14(d=56mm, h=40mm), PMA-2(d=68mm, h=61mm), and Butterfly(l1=112mm, l2=60mm, h=15mm). The second row shows the Bullet(d=20mm, l=120mm), some clutter (stone (l1=110mm, l2=56mm, h=34mm), wood cylinder (d=35mm, h=40mm), and crumpled can (l1=110mm, l2=62mm, h=15mm), and some pottery objects (perforated clay pot (d=180mm, h=28mm), clay pot (d=170mm, h=24mm), and deep clay pot (l1=120mm, l2=265mm, h=135mm)).