Unmanned Aerial Vehicle Classification via YOLOv9 and Recurrent Neural Network

Vehicle detection and classification represent critical tasks within intelligent traffic monitoring systems, yet traditional methods often pose computational challenges and limitations when adapting to varying data collection methods. This study provides a unique technique for vehicle recognition and classification in aerial image sequences. The suggested model involves six steps. Initially, all images undergo preprocessing to remove noise and raise the brightness level. At that point, foreground components are extracted using segmentation using Markov Random Field (MRF). Following segmentation, the YOLOv9 algorithm is used to detect vehicles. On the identified vehicles, feature extraction is then carried out using Binary Robust Invariant Scalable key points (BRISK), Maximally Stable Extremal Regions (MSER) and Features from Accelerated Segment Test (FAST). A genetic algorithm was employed to optimize the features. We classify via the Recurrent Neural Network (RNN) classifier. Positive results are shown by experimental results on both datasets; on the VAID dataset, the proposed model achieves an accuracy of 89.3%, and on the UAVDT, 87.2%. Furthermore, a comparative analysis with other traditional approaches is provided to evaluate the model's performance thoroughly.