Region proposal patch-image model for infrared small target detection

Gao, Chen; Zhai, Yikui

doi:10.1080/01431161.2021.1993461

Cited by 3 publications

(1 citation statement)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The detection-based deep learning methods such as regions with convolutional neural network (R-CNN) [23], Faster R-CNN [24], single shot multi-box detector (SSD) [25], and you only look once (YOLO) [26], predict infrared small targets with anchor by employing feature representation of convolutional neural network (CNN). Sommer et al [27] designed Faster R-CNN embedded region proposal network to form candidate regions for classifying and determining the infrared targets.…”

Section: B Deep Learning Methodsmentioning

confidence: 99%

Context-Guided Reverse Attention Network With Multiscale Aggregation for Infrared Small Target Detection

Zhong,

Zhang,

Duan

2023

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Infrared small target detection is one of the vital tasks in various infrared detection application, and has some typical challenges such as small and dim target, background noise, and complex scenes. To address the problem, a context guided reverse attention network (CgraNet) is proposed to detect infrared small target by introducing context guided module (CGM), multiscale aggregation block (MAB), and reverse attention module (RAM). The CGM is designed to capture inherent property of semantic information from multi-scale encode layer in pixel-level recognition. In order to eliminate the impact of low-level feature on computational complexity and ensure the detection performance, we design the MAB to aggregate multiscale feature. The RAM is integrated in decoder layer to combine the features from MAB and CGM for fusing the localization information and multiscale structural information. Extensive experiments on infrared small target datasets demonstrate that our method can achieve high detection accuracy and low false alarm rate compared with some state-of-the-art model-driven and datadriven methods.

show abstract