TY - JOUR
T1 - Cross-Layer Feature Pyramid Transformer for Small Object Detection in Aerial Images
AU - Du, Zewen
AU - Hu, Zhenjiang
AU - Zhao, Guiyu
AU - Jin, Ying
AU - Ma, Hongbin
N1 - Publisher Copyright:
© 1980-2012 IEEE.
PY - 2025
Y1 - 2025
N2 - Object detection in aerial images has always been a challenging task due to the generally small size of the objects. Most current detectors prioritize the development of new detection frameworks, often overlooking research on fundamental components such as feature pyramid networks (FPNs). In this article, we introduce the cross-layer feature pyramid transformer (CFPT), a novel upsampler-free FPN designed specifically for small object detection in aerial images. CFPT incorporates two meticulously designed attention blocks with linear computational complexity: cross-layer channelwise attention (CCA) and cross-layer spatialwise attention (CSA). CCA achieves cross-layer interaction by dividing channelwise token groups to perceive cross-layer global information along the spatial dimension, while CSA enables cross-layer interaction by dividing spatialwise token groups to perceive cross-layer global information along the channel dimension. By integrating these modules, CFPT enables efficient cross-layer interaction in a single step, thereby avoiding the semantic gap and information loss associated with elementwise summation and layer-by-layer transmission. In addition, CFPT incorporates global contextual information, which improves detection performance for small objects. To further enhance location awareness during cross-layer interaction, we propose the cross-layer consistent relative positional encoding (CCPE) based on interlayer mutual receptive fields. We evaluate the effectiveness of CFPT on three challenging object detection datasets in aerial images: VisDrone2019-DET, TinyPerson, and xView. Extensive experiments demonstrate that CFPT outperforms state-of-the-art FPNs while incurring lower computational costs.
AB - Object detection in aerial images has always been a challenging task due to the generally small size of the objects. Most current detectors prioritize the development of new detection frameworks, often overlooking research on fundamental components such as feature pyramid networks (FPNs). In this article, we introduce the cross-layer feature pyramid transformer (CFPT), a novel upsampler-free FPN designed specifically for small object detection in aerial images. CFPT incorporates two meticulously designed attention blocks with linear computational complexity: cross-layer channelwise attention (CCA) and cross-layer spatialwise attention (CSA). CCA achieves cross-layer interaction by dividing channelwise token groups to perceive cross-layer global information along the spatial dimension, while CSA enables cross-layer interaction by dividing spatialwise token groups to perceive cross-layer global information along the channel dimension. By integrating these modules, CFPT enables efficient cross-layer interaction in a single step, thereby avoiding the semantic gap and information loss associated with elementwise summation and layer-by-layer transmission. In addition, CFPT incorporates global contextual information, which improves detection performance for small objects. To further enhance location awareness during cross-layer interaction, we propose the cross-layer consistent relative positional encoding (CCPE) based on interlayer mutual receptive fields. We evaluate the effectiveness of CFPT on three challenging object detection datasets in aerial images: VisDrone2019-DET, TinyPerson, and xView. Extensive experiments demonstrate that CFPT outperforms state-of-the-art FPNs while incurring lower computational costs.
KW - Aerial image
KW - feature pyramid network (FPN)
KW - object detection
KW - vision transformer (ViT)
UR - https://www.scopus.com/pages/publications/105006579283
U2 - 10.1109/TGRS.2025.3572706
DO - 10.1109/TGRS.2025.3572706
M3 - Article
AN - SCOPUS:105006579283
SN - 0196-2892
VL - 63
JO - IEEE Transactions on Geoscience and Remote Sensing
JF - IEEE Transactions on Geoscience and Remote Sensing
M1 - 5625714
ER -