Autonomous UAV path planning and landmine risk mapping using SLAM-integrated real-time thermal-RGB vision sensing

Landmine contamination remains a critical threat in post-conflict zones, necessitating rapid, accurate, and autonomous surveying solutions. This paper presents a new Deep Convolutional Neural Network (DCNN) framework in identifying landmines under the ground with aided aerial RGB and thermal image data taken by Unmanned Aerial Vehicles (UAVs). The model bypasses the often-required role of direct human interaction and has fewer demands on the direct ability to generate detections to autonomously detect buried landmines. The suggested system is based on a Dual-Modality Cross-Attention Fusion (DCAF) module to merge the RGB and the thermal feature. The model attained the overall accuracy of 97.89 %, precision of 97.4 %, recall of 98.1 %, and F1-score of 97.7 % on a variety of simulated terrains, such as grasslands, muddy areas, rubbles, and cratered disposals. Also, the system shows a temporal consistency index of 96.5 % and a mean IoU of 93.1 % surpassing the RGB-only and thermal-only baselines.