RINKA: efficient artificial neural network model for drone-based object detection
Fragment książki (Rozdział monografii pokonferencyjnej)
MNiSW
20
Poziom I
| Status: | |
| Autorzy: | Tomiło Paweł |
| Dyscypliny: | |
| Aby zobaczyć szczegóły należy się zalogować. | |
| Wersja dokumentu: | Drukowana | Elektroniczna |
| Język: | angielski |
| Strony: | 209 - 216 |
| Web of Science® Times Cited: | 0 |
| Bazy: | Web of Science |
| Efekt badań statutowych | NIE |
| Materiał konferencyjny: | TAK |
| Nazwa konferencji: | 8th International Conference on Image, Video and Signal Processing |
| Skrócona nazwa konferencji: | 8th IVSP 2026 |
| URL serii konferencji: | LINK |
| Termin konferencji: | 17 marca 2026 do 19 marca 2026 |
| Miasto konferencji: | Tokyo |
| Państwo konferencji: | JAPONIA |
| Publikacja OA: | NIE |
| Abstrakty: | angielski |
| With the rapid development of unmanned aerial vehicle (UAV) technology, drone-based aerial imaging and remote sensing have gained importance as effective tools in civil and military applications. Despite their many advantages, object detection in images acquired from UAVs remains a challenge due to scale variability, complex backgrounds, low object resolution, and varying environmental conditions. In response to these difficulties, this paper proposes a new object detection model, RINKA (Repeated efficient layer aggregation network, INvolution, Kolmogorov-Arnold), which combines the Involution mechanism with an architecture based on the Kolmogorov-Arnold network. This model was designed with adaptability to local image features and high computational efficiency in mind. As part of the experiments, the RINKA model was compared with modern detection architectures from the YOLO family (YOLOv8. YOLOv9, YOLOv10, YOLOv11). The results showed that RINKA achieved the highest evaluation metrics, confirming its effectiveness in diverse conditions. |
