2D-GAN: Dual-Decoding Generative Adversarial Networks for Infrared Image Enhancement

Infrared imaging technology finds wide application across various fields, yet suffers from issues such as atmospheric thermal radiation interference, resulting in poor contrast, blurred details, and noise in infrared images. Consequently, enhancing infrared images is essential as a preprocessing step to obtain high-quality data. This paper proposes an infrared image enhancement algorithm based on Dual-Decoding Generative Adversarial Networks(2D-GAN) to address these challenges.The proposed algorithm employs a two-step decoding structure within the 2D-GAN framework to enhance the network's capability to comprehend and represent input data effectively. Internal and external skip connections are incorporated to bolster the network's perceptual ability, addressing the loss of detailed information during the encoding and decoding processes. Additionally, a cross-level attention module is designed to dynamically allocate positional weights to feature maps, thereby enhancing the natural appearance of the generated images.The effectiveness of the 2D-GAN-based enhancement algorithm is validated through comparative experiments, supplemented by ablation studies that delineate the contributions of each module within the network. Experimental results demonstrate significant enhancement in infrared image quality, affirming the efficacy of the proposed algorithm.