.png){kind=icon}
.png){kind=icon}
Return
Enhancing small object detection: LDNet with location awareness and detail enhancement
.png){kind=icon}
PRE
AI
C
H
W
B
DOI：10.1016/j.patrec.2026.03.012.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Small object detection is a challenging task in computer vision because small objects are difficult to locate and have limited detailed information. Although recent methods have improved spatial information perception through attention mechanisms, the standard convolution operations they rely on still struggle to adaptively capture location of small objects. In addition, while multi-scale feature fusion alleviates the problem of insufficient discriminative features for small objects, existing feature pyramid networks remain limited in their ability to enhance details. To address these issues, we propose LDNet, a novel small object detection network that leverages location awareness and detail enhancement. LDNet introduces a dynamic location awareness module to strengthen spatial information perception and a detail-enhanced feature pyramid network to improve feature representation. Experimental results on the VisDrone and FBD-SV-2024 datasets demonstrate the effectiveness of LDNet, achieving average precision improvements of 1.3% and 6.1% over the baseline QueryDet, respectively. These results validate the superior performance of LDNet in accurately detecting small objects in complex scenes. The source code is available at: https://github.com/judycpChen/LDNet.
Keywords:
Small object detection
Location awareness
Detail enhancement
Feature pyramid network
Dynamic convolution
Journal
IF:
3.3
Papers:
7.8K
Citations:
1.6W
Organization
Cited Papers