Data generation for nighttime infrared scenes using enhanced domain migration learning

In nighttime infrared scenarios, data generation plays a pivotal role. However, the limited availability of nighttime infrared image datasets often constrains the model's generalization ability and accuracy. To mitigate this challenge, this study proposes a novel method for generating infrared image data by integrating adversarial training techniques with foreground enhancement methods. A foreground enhancement algorithm is employed to process visible light images, thereby enhancing the sensitivity of visible light images to targets during the CYCLEGAN model generation process. This approach effectively improves the quality of the generated nighttime infrared images and enhances the performance of target detection in such images, resulting in notable enhancements. Experimental results on publicly available infrared image datasets demonstrate that training with augmented data can significantly improve the performance of target detection models, providing a new solution for scenarios with limited nighttime infrared image data.