.png){kind=icon}
.png){kind=icon}
Return
Infrared target detection performance enhancement using TSR-LMS algorithm
.png){kind=icon}
OA
AI
DOI：10.1016/j.infrared.2025.106178.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Infrared weak target detection is one of the key technologies in infrared search and tracking. It is widely used in maritime surveillance systems and reconnaissance systems, with significant potential for civilian applications such as security monitoring, forest fire detection, and power equipment inspection. However, detecting infrared weak targets in low-resolution images faces challenges such as small target size, sparse texture information, complex backgrounds, and low signal-to-noise ratio(SNR). Inspired by the positive impact of super-resolution on target detection, we introduced an improved version of the classic Temporally Selective Regularized Least Mean Squares (TSR-LMS) video super-resolution algorithm to enhance target detection performance. To further reduce noise introduced by registration errors, we employ the Demons registration algorithm to address nonlinear transformations between adjacent frames. The effectiveness of our algorithm is validated on publicly available infrared target detection datasets and real image sequences captured by infrared cameras. Experimental results demonstrate that the method effectively suppresses background clutter interference while improving target signal-to-noise ratio. To further validate the effectiveness of the algorithm, we selected several representative infrared small target detection algorithms and performed target detection on image sequences from multiple different scenarios, comparing the detection results before and after reconstruction. The results demonstrate a significant improvement in the average AUC of the reconstructed images across image sequences from various scenarios. Thus, our method significantly enhances the detection probability of traditional single-frame and multi-frame target detection algorithms while reducing their false alarm rate, demonstrating broad application prospects in infrared target detection.
Keywords:
Target detection
Video super-resolution reconstruction
TSR-LMS
Demons registration
TSR-LMS
Temporally Selective Regularized Least Mean Squares
AUC
The average area under the curve
STLCF
Spatial–Temporal local contrast filter
STLCM
The spatial–temporal local contrast map
STP
Spatiotemporal saliency model and prediction method
LR
Low-resolution
HR
High-resolution
R-LMS
The recursive least mean squares
SNR
Signal-to-noise ratio
SCR
Signal-to-clutter ratio
MDSR
The Multi-scale Deep SuperResolution
GSD
Ground sample distance
AP
The average precision
SRCNN
Super-resolution convolutional neural network
GFLOPS
Giga Floating Point Operations Per Second
CNN
Convolutional Neural Network
RLCM
The Relative Local Contrast Measure
MPCM
The multiscale patch-based contrast measure
EDVR
Video Restoration with Enhanced Deformable Convolutional Networks
VRT
Video super-resolution Transformers
ROC
Receiver Operating Characteristic
ABC
Artificial bee colony
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
I
IF:
3.4
Papers:
5.8K
Citations:
1.2W