.png){kind=icon}
.png){kind=icon}
Return
LCS-YOLO: a lightweight YOLO11n algorithm for infrared small target detection
.png){kind=icon}
PRE
AI
H
H
H
Y
Z
T
DOI：10.1117/1.JEI.35.2.023038.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
To address the problems of missed detections and false detections frequently occurring in traditional target recognition algorithms due to the small target size, low image resolution, and complex background of infrared images, a Lightweight infrared small target detection algorithm based on YOLO11n, named LCS-YOLO (Lightweight Cross-Scale dynamic attention YOLO), is proposed. First, a lightweight multiscale feature attention module is introduced into the backbone network to fuse multiscale features while significantly reducing computational cost. Second, a P2 small target detection layer is added to the neck network, and a weighted bi-directional feature pyramid network is employed to extract features at different scales. Then, a lightweight channel-dynamic feature attention module is applied to reconstruct the C2f convolution. Finally, although light-weighting the entire network structure, a dynamic detection head (SE-Dynamic Head) is designed to enhance the detection performance of infrared small targets under complex backgrounds and reduce computational cost. Compared with the YOLO11n model, the proposed LCS-YOLO algorithm improves mAP@50 by 5.4%, with only 0.22 M trainable parameters. Moreover, it has also achieved significant improvements in accuracy and recall rates. It is suitable for real-time infrared small target detection tasks. (C) 2026 SPIE and IS&T
Keywords:
target detection
infrared small target
lightweight feature module
weighted feature fusion
multi-scale feature extraction
dynamic attention mechanism
Journal
J
IF:
1
Papers:
109
Citations:
2.7K
Organization
Cited Papers