Hierarchical Feature Fusion for Multi-scale Target Region Detection of Fixed-Wing UAVs

This paper studies the problem of fixed-wing unmanned aerial vehicle (UAV) target-region detection. This problem is important for applications such as smalltarget detection, vision-based UAV swarming, and fault detection. To address deficiencies in feature-region detection datasets and algorithms for fixed-wing UAVs, we construct a fixed-wing UAV feature-region detection dataset and propose a hierarchical feature fusion-based multi-scale feature-region detection algorithm. This paper explores methods for simulated fixed-wing UAV image collection and automatic data annotation. A comprehensive dataset encompassing simulated scenarios is developed. In total, the dataset includes 108,000 images with corresponding annotations. Furthermore, on the multi-scene simulated dataset, the proposed algorithm achieves a detection accuracy of 70.44%, representing an average improvement of 5.07% over baseline algorithms. The results highlight key challenges in fixed-wing detection and indicate possible directions for future algorithm development.