.png){kind=icon}
Return
Multi-scale pyramid convolution transformer for remote-sensing object detection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.imavis.2025.105651.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• A novel pyramid convolution Transformer framework for RSOD is proposed, which uses dual-route branch in each encoder block to enhance essential inductive bias for multi-scale object, thereby improving discriminative feature representation capability. • To deal with large-scale variation, bidirectional feature aggregation is devised to facilitate deep interaction of high-level semantic features and low-level detail information, enhancing the multi-scale object representation capability of the network. • To reduce computation cost, a sparse sampling scheme is introduced by adjusting the encoder-decoder connection relationship, which iteratively identifies regions of interest under the guidance of prior detection results, then extracts multi-scale features at each sparse keypoint. • Comprehensive experiments are conducted to confirm the validity of the proposed MPCViT. The performance gains of 1.29% and 2.13% considerably outperform the state-of-the-art ARS-DETR on NWPU VHR-10 and DIOR datasets respectively.
Journal
IF:
4.2
Papers:
4.1K
Citations:
6.7K
Organization
No organization information available
Cited Papers
.png){kind=icon}
ShareSaveShareSaveShareSave
View more