UAV-MS: Anti-Interference Data Fusion Algorithm for UAV Multimodal Sensors in the Internet of Things

This study proposes an anti-interference data fusion algorithm called UAV-MS, aimed at solving the data fusion problem of multimodal sensors in unmanned aerial vehicle (UAV) applications. With the widespread use of UAVs in fields such as agricultural monitoring and environmental detection, the demand for improving their autonomy and accuracy is growing. UAV-MS enhances the system's adaptability to complex environments by integrating data from various sensors, including visual images, LiDAR point clouds, and infrared thermal imaging. In the face of challenges where traditional methods struggle to handle the inconsistency between heterogeneous data sources, this study utilizes deep learning techniques to design an innovative data fusion algorithm. The algorithm is specifically tailored for eight key sensors, including the Sony IMX series cameras and Velodyne VLP-16 LiDAR. It employs a model structure consisting of three parts: feature extraction, denoising network, and fusion network. The feature extraction phase uses spatiotemporal convolution and the MLPMixer mechanism; the denoising network, based on a GAN architecture, effectively handles Gaussian noise of varying intensities; while the fusion network combines LSTM and Transformer blocks to achieve effective integration of sequential and spatial features. Experimental verification shows that UAV-MS significantly improves UAV performance in various real-world scenarios, outperforming other models with higher accuracy in both 10-class and 5-class tasks, providing strong technical support for the development of intelligent UAVs.