Open AccessInfrared Small Target Detection Based on Multi-Scale Local Contrast Network
Author(s) -
Yuanteng Liu,
Yuehuan Wang
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1883/1/012077
Subject(s) - contrast (vision) , computer science , fuse (electrical) , scale (ratio) , artificial intelligence , computation , pattern recognition (psychology) , layer (electronics) , frame (networking) , pixel , fusion , computer vision , algorithm , telecommunications , engineering , geography , cartography , linguistics , chemistry , philosophy , organic chemistry , electrical engineering
To mitigate the challenge of the lack of the intrinsic features of the small target which possesses only few pixels, we designed a multi-scale local contrast network which combines features extracted from different scales and layers for single-frame infrared small target detection in this paper. We extract the multi-scale contrasts features in the same layer by multi-scale local contrast module and fuse the cross-layer contrast features by local contrast fusion module. Moreover, we embed these modules at each stage as a depth-wise nonlinear layer into an end-to-end convolutional network to predict the small target. We conduct extensive experiments and comparison with other SOTA methods to verify the performance of our method. The results show that our method outperforms many of its competitors, and when taking computation cost into consideration, it will be more outstanding.