Advances and Emerging Trends in UAV Sensor Data Fusion Technology

With the acceleration of large-scale application of UAV in agricultural monitoring, disaster rescue, logistics and distribution and other fields, its complex environmental perception bottleneck is becoming increasingly prominent. The traditional single sensor system is limited in GNSS signal shading, IMU cumulative drift and visual photosensitivity, which seriously restricts the system reliability and adaptability of the system. This paper systematically reviews the breakthrough progress of multi-source data fusion technology: the space-time fusion architecture based on the Kalman filter (UKF) reduces the dynamic attitude estimation error by 18%, and the GPU accelerated particle filtering scheme realizes 20 Hz real-time processing in real-time. The proposed precision-cost-power consumption triangle constraint criterion guides the combination of multispectrum and RTK-GNSS to balance 95% classification accuracy with 10 cm localization error in agricultural monitoring. Typical application verification shows that the bionic radar-vision system achieves 0.32 m positioning accuracy in the occlusion environment (41% improvement compared with the traditional method), and the LiDAR-photogrammetry combined leveling technology improves the absolute accuracy of terrain modeling by 23%. The multi- dimensional evaluation system constructed in this study (localization error of 5 cm, response delay of <100 ms, work15 W) provides theoretical support and decision basis for engineering deployment under complex working conditions.