UAV-Borne Landmine Detection via Intelligent Multispectral Fusion

Compared with traditional landmine detection methods, the unmanned aerial vehicle borne measure is more suitable for rapidly detecting large-area minefields of scatterable landmines. We propose a multispectral fusion strategy based on a deep learning model to facilitate landmine detection. With a UAV-borne multispectral cruise platform, we set up a multispectral dataset of scatterable landmines, where varieties of minefields with different ground vegetation environments are considered. The proposed technique includes a structure with adjustable weights to fulfill the fusion and refinement, and maximize the utilization of multispectral information. Experimental comparisons of the predictions before and after fusion demonstrate that our proposed fusion strategy can improve the recall rate and achieve significant gains over single-spectrum detection, especially for the occluded mines, and confirm the feasibility of the multispectral data with reasonable processing for the landmine detection.