Autonomous Landing of Unmanned Aerial Vehicles: Hybrid Metaheuristic Aided Detection and Extended Kalman Filter for Localization

The majority of uses for UAVs are in precision agriculture, cinematography, aerial surveillance, military applications, disaster relief, construction inspection, and other applications. Nevertheless, a precise localization method is needed to accomplish these tasks on their own. This study concentrated on autonomous ground vision-guided landing of a fixed-wing unmanned aerial vehicle (UAV) in situations where GNSS is unavailable. To improve image detection and accuracy for UAV auto-landing, optimum Deep Learning (DL) approaches would be used in the design and implementation. Therefore, this research presents a DL-based auto-landing scheme for UAVs. Once the flying vehicle's motion consistency is confirmed to be unsatisfactory, an optimized Long Term Short Memory (LSTM) would be constructed to improve detection accuracy. To adjust the weight of the LSTM, an RDSBWO (Red Deer struck Black widow optimization) technique should be employed. The normal BWO and RDA are conceptually combined to create the RDSBWO model. The proposed DL concept uses image detection and spatial placement to automatically improve the inaccurate coordinates within a given range. Moreover, object tracking and localization were accomplished using Extended Kalman Filter (EKF-based) object localization. Comparing the adopted established LSTM + RDSBWO scheme to existing approaches such as LSTM + BWO (0.36), LSTM + RDA (0.439), LSTM + LA (0.489), LSTM + BOA (0.690), LSTM + AOA (0.489), LSTM + PRO (0.389), GRU (0.489), DBN (similar to 0.610), and LSTM (similar to 0.690), it achieves lower MSE (0.235) with better performance for the number of frames 4.