Comparative Analysis of Ground Target Detection Algorithms Based on XTDrone

In recent years, with the rapid development of drone technology, unmanned aerial vehicles (UAVs) have been widely applied in various fields such as terrain exploration and aerial photography. Ground target detection is one of the many application areas of UAVs. Through ground target detection, UAVs are able to quickly obtain information about the position, quantity, and size of targets without physical contact with the ground, providing data support for various application scenarios. Therefore, this paper conducts experimental analysis based on the XTDrone simulation platform, combined with target detection algorithms YOLOv4, YOLOv4-tiny, and YOLOv7. By comparing the similarities and differences among the algorithms, the impact on the final training results is analyzed. Additionally, an analysis of the settings of hyperparameters is performed, and it is found that the learning rate has a significant influence on the training results, while the batch size does not have a significant effect. Furthermore, the size of the dataset is identified as an important factor restricting the training effectiveness. Finally, this paper presents a visualization demonstration of the training results using YOLOv4-tiny as an example, showing good recognition performance that aligns with expectations.