一种双分支特征交互融合的高效红外图像彩色化方法

打开文本图片集

中图分类号：TN911.73；TP391文献标志码：ADOI：10.7652/xjtuxb202508020 文章编号：0253-987X（2025）08-0211-12

An Efficient Infrared Image Colorization Method Based on Dual-Branch Feature Interaction Fusion

CHEN Yu1， ZHAN Weida' ，JIANG Yichun'， ZHU Depeng1 ，HAN Deng1， （1.Electronicand Information Engineering，Changchun Universityof Science and Technology，Changchun l30022，China 2. Jilin Province Zhixing IoT Research Institute Co.，Ltd.，Changchun ，China）

Abstract： To address the significant limitations of existing infrared image colorization methods in global feature capture and computational complexity，an eficient infrared image colorization method based on dual-branch feature interaction fusion is proposed. A dual-branch encoder is designed，where the local feature extraction branch captures local spatial context information to ensure fine-grained feature acquisition， while the global feature extraction branch obtains global features to meet long-range dependency requirements. An interaction fusion module is developed to effectively integrate features extracted from both branches，significantly enhancing the model' s overall performance. In the decoder part，a context aggregation module is proposed to further optimize multi-scale semantic feature aggregation，improving edge clarity and detail representation in the colorization results. Extensive experimental validation on the KAIST and FLIR datasets demonstrates that compared to existing methods，the proposed approach achieves superior colorization quality on both datasets，with peak signal-to-noise ratios reaching 28.645 dB and 30，459dB ，and structural similarity scores of O.507 and 0.725，respectively，outperforming all comparative methods. The effectiveness and advancement of the method are thus verified. The research findings provide valuable references for enhancing the readability and interpretability of infrared images，as well as improving observation capabilities in night vision and harsh environments.

Keywords： infrared image colorization； fine-grained feature； long-range dependency； interactivefusion; context aggregation

红外图像因其能够捕捉人眼无法感知的热辐射信息，在弱光、烟雾和恶劣天气等复杂环境中，尤其在目标检测、识别与追踪等方面展现了独特的优势[1]。（剩余18543字）