We propose a magnetometer system fitted on an unmanned aerial vehicle (UAV, or drone) and a data-processing method for detecting metal antipersonnel landmines (M16) in the demilitarized zone (DMZ) in Korea, which is an undeveloped natural environment. The performance of the laser altimeter was improved so that the drone could fly at a low and stable altitude, even in a natural environment with dust and bushes, and a magnetometer was installed on a pendulum to minimize the effects of magnetic noise and vibration from the drone. At a flight altitude of 1 m, the criterion for M16 is 5 nT. Simple low-pass filtering eliminates magnetic swing noise due to pendulum motion, and the moving average method eliminates changes related to the heading of the magnetometer. Magnetic exploration was conducted in an actual mine-removal area near the DMZ in Korea, with nine magnetic anomalies of more than 5 nT detected and a variety of metallic substances found within a 1-m radius of each detection site. The proposed UAV-based landmine detection system is expected to reduce risk to detection personnel and shorten the landmine-detection period by providing accurate scientific information about the detection area prior to military landmine-detection efforts.

Many antipersonnel (AP) and antitank (AT) mines remain in the ground across South Korea; some are partially exposed and remain visible. We used a drone fit with a magnetometer to detect landmines. The magnetic characteristics of landmines were measured, and altitude and flight conditions (altitude <1 m) required for successful landmine detection were derived. Metallic AP and AT mines, and low metal content AT mines were detected in field tests under the proposed conditions. The drone can be used to aid in the demining of South Korea.

Side-scan sonar (SSS)-based detection of unexploded ordnance is challenging because complex seabed patterns, variable acoustic-shadow geometry, and nonuniform background backscatter often lead to both false positive (FP) and false negative (FN) detections. Conventional annotation schemes often represent an object and its acoustic shadow as a single region, making it difficult to explicitly capture their structural relationship and potentially including unnecessary background areas. This study proposes a shadow-extended oriented bounding box (SE-OBB) annotation scheme to explicitly represent the spatial relationship between an object and its acoustic shadow along the across-track range direction in SSS imagery. SE-OBB employs a dual-class structure in which Class 0 represents the object region and Class 1 represents an extended region containing both the object and its acoustic shadow. The effects of overlapping-mask processing were also evaluated by comparing the overlap_mask=True/False conditions. To validate the proposed approach, datasets annotated using bounding box, OBB, SE-OBB, and Segmentation were converted into training masks and used to fine-tune YOLO26s under identical Segmentation Task conditions. During inference, initial candidates were detected using a multi-scale sliding-window strategy, followed by a four-stage post-processing pipeline based on sonar physical characteristics and object-shadow structural relationships. At an intersection over union (IoU) threshold of 0.5, SE-OBB with overlap_mask=True and the proposed post-processing pipeline achieved the best overall performance among all experimental conditions, with a precision of 0.715, recall of 0.844, F1 of 0.774, and F2 of 0.815. Compared with the baseline SE-OBB detections, post-processing reduced FP from 98 to 41, corresponding to a 58.2% decrease, while true positive increased from 95 to 103 and FN decreased from 27 to 19. These results demonstrate that SE-OBB can effectively represent object-shadow structures through a simpler annotation procedure than Segmentation and, when combined with sonar-physics-based post-processing, can mitigate the precision-recall trade-off without additional model retraining.