PATR16: a dataset for aircraft type recognition in remote sensing images

Currently, the development of datasets for fine-grained recognition of aircraft targets in remote sensing images remains insufficient, with existing datasets often covering aircraft types too broadly and lacking specificity. To address this issue, this study focuses on common aircraft models deployed at major airports in the Asia-Pacific region. Utilizing highresolution remote sensing imagery as the primary data source, supplemented by historical imagery, global airport imagery, and original images from related datasets, we extensively collected relevant images and annotated the targets within them. This process resulted in a complete dataset with standardized content and format, named the PATR16 (Remote Sensing Image Aircraft Type Recognition Dataset).Subsequently, two deep learning-based object detection algorithms, Faster R-CNN and YOLOv11, were employed to train and test the dataset. Experimental results demonstrate that Faster R-CNN achieves higher average recognition accuracy than YOLOv11, with the mean accuracy for most aircraft models exceeding 80%, enabling reliable identification of the relevant aircraft types. The dataset constructed in this study effectively supports the recognition of target aircraft models and contributes to advancing fine-grained interpretation techniques for remote sensing image target identification.