Comparing Surface Landmine Object Detection Models on a New Drone Flyby Dataset

Navin Agrawal-Chung,Zohran Moin

2024-10-17

Abstract:Landmine detection using traditional methods is slow, dangerous and prohibitively expensive. Using deep learning-based object detection algorithms drone videos is promising but has multiple challenges due to the small, soda-can size of recently prevalent surface landmines. The literature currently lacks scientific evaluation of optimal ML models for this problem since most object detection research focuses on analysis of ground video surveillance images. In order to help train comprehensive models and drive research for surface landmine detection, we first create a custom dataset comprising drone images of POM-2 and POM-3 Russian surface landmines. Using this dataset, we train, test and compare 4 different computer vision foundation models YOLOF, DETR, Sparse-RCNN and VFNet. Generally, all 4 detectors do well with YOLOF outperforming other models with a mAP score of 0.89 while DETR, VFNET and Sparse-RCNN mAP scores are all around 0.82 for drone images taken from 10m AGL. YOLOF is also quicker to train consuming 56min of training time on a Nvidia V100 compute cluster. Finally, this research contributes landmine image, video datasets and model Jupyter notebooks at <a class="link-external link-https" href="https://github.com/UnVeilX/" rel="external noopener nofollow">this https URL</a> to enable future research in surface landmine detection.

Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is: **How to use deep learning and UAV (unmanned aerial vehicle) videos to detect surface landmines, especially small - sized surface landmines (such as POM - 2 and POM - 3), efficiently and safely**. Traditional methods (such as manual metal detectors and armored mine - clearing vehicles) have the problems of being slow, dangerous and costly. Especially in areas such as Ukraine, due to the wide and rapid distribution of landmines, there is an urgent need for faster and safer detection techniques.

Specifically, the paper mainly focuses on the following aspects:

1. **Create a UAV image data set suitable for surface landmine detection**: Existing public data sets lack data for surface landmines. Therefore, researchers have created a data set containing UAV images and videos under different environmental conditions for training and testing models.

2. **Evaluate the performance of multiple deep - learning object - detection models in landmine detection**: The paper selected four different computer - vision - based models (YOLOF, DETR, Sparse - RCNN and VFNet), and trained, tested and compared these models on the self - built data set. The focus was on evaluating their accuracy and inference speed.

3. **Explore the application potential of UAV videos in landmine detection**: Through experiments, the detection effects of UAVs at different flight altitudes (2.5m, 5m and 10m) were verified, and the performance differences of different models on high - resolution images were analyzed.

4. **Provide resources and support for future research**: The paper not only provides trained models and data sets, but also shares relevant Jupyter notebooks, so that other researchers can further carry out work on this basis.

In summary, this paper aims to develop an efficient, safe and economically viable surface landmine detection method through deep learning and UAV technology to meet the challenges currently faced in landmine clearance.