Résumé: Infrared images are full of energy information, which can intuitively reflect the difference between objects and scenes. Visible images are full of color information, which can intuitively reflect the details of objects and scenes. To achieve the comprehensive utilization of infrared energy information when transforming infrared images to visible images, this paper proposes a method called infrared-energy-to-color (IETC) based on Cycle-Consistent Generative Adversarial Networks (CycleGAN). We compare some state-of-the-art methods in image transformation; the results show that the proposed model can generate visible images of ocean objects and scenes in the visual expression. Furthermore, in the quantitative evaluation, the proposed method performs better than others. In addition, the classic object detection algorithms also show that the proposed IETC method can achieve more robust results.

Langue: Anglais