Unmanned aerial vehicle terrain matching algorithm based on multimodal feature fusion and particle swarm optimization

Autonomous navigation for Unmanned Aerial Vehicles (UAVs) is critical in Global Navigation Satellite System (GNSS)-denied scenarios, particularly within complex electromagnetic environments. Conventional Terrain Aided Navigation (TAN) systems often rely on single-modality sensors, making them susceptible to targeted interference that can degrade feature data and lead to positioning failure. Although multimodal feature fusion has shown potential for enhancing robustness, existing methods often impose significant computational overhead, limiting their suitability for real-time UAV applications.

This study aims to develop a robust and computationally efficient terrain matching algorithm that enhances resilience against electromagnetic interference, mitigates fusion bias caused by disparate feature scales, and improves search efficiency to meet real-time operational requirements.

The proposed algorithm integrates a dual-modality feature fusion framework. Rotation Invariant Uniform Local Binary Pattern (RIULBP) features are extracted from Synthetic Aperture Radar (SAR) imagery to capture noise-resistant spatial textures, while Frequency Energy Distribution (FED) features are derived from Digital Elevation Models (DEM) to represent global terrain structure. A dynamic weighting method based on feature sensitivity is employed to fuse these heterogeneous features, with Z-score normalization used to standardize their scales. The fused Canberra distance serves as the similarity metric for terrain matching. Particle Swarm Optimization (PSO) replaces the conventional sliding-window search, enabling efficient identification of the optimal match within the search area.

Experimental evaluations on a diverse dataset, including mountains, plains, and deserts, demonstrated that the proposed algorithm achieved a matching success rate consistently above 90%, outperforming single-modality and fixed-weight fusion methods. The algorithm also exhibited strong robustness in anti-interference tests, where Gaussian, speckle, and impulse noise were injected into SAR images, achieving up to a 30% improvement in matching success rate compared to single-modality approaches. Additionally, the PSO-based search significantly reduced computational time compared to exhaustive search methods.

The proposed algorithm provides an effective solution for UAV autonomous navigation in challenging environments. By combining spatial-domain (RIULBP) and frequency-domain (FED) features through a dynamic weighting strategy, the algorithm enhances robustness against electromagnetic interference while maintaining computational efficiency. The integration of PSO further ensures real-time applicability, validating the effectiveness of multimodal fusion and intelligent optimization for reliable UAV positioning.