Title: RCAU-Net: convolutional networks with residual channel attention for non-uniformity correction

Authors: Feng Deng; Shiqiang Chen; Yutian Ma; Shaoyi Cheng; Yuanjun Sun; Jingjing Yang

Addresses: Beijing Key Laboratory of High Dynamic Navigation Technology, Beijing Information Science and Technology University, Beijing, 100192, China; Key Laboratory of Modern Measurement and Control Technology, Ministry of Education, Beijing Information Science and Technology University, Beijing, 100192, China; School of Automation, Beijing Information Science and Technology University, Beijing, 100192, China ' Beijing Key Laboratory of High Dynamic Navigation Technology, Beijing Information Science and Technology University, Beijing, 100192, China; Key Laboratory of Modern Measurement and Control Technology, Ministry of Education, Beijing Information Science and Technology University, Beijing, 100192, China; School of Automation, Beijing Information Science and Technology University, Beijing, 100192, China ' National Key Laboratory of Ni&Co Associated Minerals Resources Development and Comprehensive Utilization, Jinchang, 737104, China ' National Key Laboratory of Ni&Co Associated Minerals Resources Development and Comprehensive Utilization, Jinchang, 737104, China ' National Key Laboratory of Ni&Co Associated Minerals Resources Development and Comprehensive Utilization, Jinchang, 737104, China ' Beijing Key Laboratory of High Dynamic Navigation Technology, Beijing Information Science and Technology University, Beijing, 100192, China; School of Automation, Beijing Information Science and Technology University, Beijing, 100192, China

Abstract: Infrared imaging technology takes on critical significance in surveillance and security. However, the non-uniformity present in the infrared detectors manifests mainly as stripe noise in infrared images, severely limiting the sensitivity and advancement of the infrared imaging system. Most existing non-uniformity correction methods suffer from redundant noise and incomplete preservation of image details. This study introduces a novel correction model for non-uniformity grounded in the U-Net framework. This model incorporates a deep residual network, a channel attention mechanism, and global residual learning with the U-Net encoder- decoder for effective stripe noise reduction, which facilitates the learning of more profound context features and enables more accurate extraction of stripe noise. Our proposed approach has undergone evaluation using both simulated and real-world image data, revealing promising results through comparative analysis. Compared with the U-Net model, our model has shown improvements of 1.26 dB in peak signal-to-noise ratio and 0.7% in structural similarity.

Keywords: non-uniformity correction; NUC; stripe noise; channel attention; deep residual network; residual learning.

DOI: 10.1504/IJSNET.2024.137337

International Journal of Sensor Networks, 2024 Vol.44 No.3, pp.169 - 181

Received: 14 Oct 2023
Accepted: 19 Oct 2023
Published online: 12 Mar 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article