UAV image object recognition method based on small sample learning

• Published in: Multimedia tools and applications Vol. 82; no. 17; pp. 26631 - 26642
• Main Authors: Tan, Li, Lv, Xinyue, Wang, Ge, Lian, Xiaofeng
• Format: Journal Article
• Language: English
• Published: New York Springer US 01-07-2023; Springer Nature B.V
• Subjects: Aerial photography; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Learning; Multimedia Information Systems; Object recognition; Special Purpose and Application-Based Systems; Target recognition; Unmanned aerial vehicles; Small sample learning; Feature fusion; UAV; Object recognition; Data enhancement
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-14985-y

In recent years, unmanned aerial vehicles (UAVs) have developed rapidly. Because of their small size, low cost, and strong maneuverability, they have been widely used in several fields such as aerial photography, rescue, transportation, and agriculture. Object recognition requires a large amount of data, but in real application scenarios, due to factors such as privacy and high data labeling costs, it is impossible to obtain sufficient label training samples. This paper proposes an unmanned aerial vehicle (UAV) image object recognition model based on small sample learning (IORS). Based on data enhancement and improved feature fusion capabilities, the YOLOv4_Tiny model is improved to make it more applicable to UAV images. This solves the problem of identifying dense small targets in UAV images when dealing with a small number of samples. The experimental results showed that in UAV images, the proposed method has a good target recognition effect without reducing the speed, while the overall accuracy is increased by approximately 4.5%.