To enhance the environmental perception and target recognition capabilities of unmanned aerial vehicles (UAVs) in complex environments, the exploration of UAVs-mounted day-and-night single-channel natural-sensing color thermal imaging technology was undertaken. Based on deep learning technology, two image colorization techniques were constructed: the TIVNet, an infrared image colorization network based on a multi-discriminator generative adversarial network, and the infrared image colorization network based on the semantic-guided diffusion model with regional self-segmentation RSDM. A generative adversarial network architecture with multi-discriminator was employed by TIVNet to directly convert infrared thermal images into colorized visible-light-like images. However, there were instances where color inaccuracies were observed in the detailed aspects of certain scenes. The more real color images through a more sophisticated model were generated by RSDM, enhancing the visual effect and efficiency of information transmission in images, but the processing speed of the current model needed to be improved. Experimental results indicate that the two proposed methods each possesses distinct advantages in terms of the fidelity of image conversion and real-time processing rate. TIVNet has achieved real-time processing at a rate of not less than 40 Hz on platforms such as UAVs, demonstrating the technical feasibility. Moreover, it possesses the capability to facilitate equipment transformation applications based on the specific application requirements, thereby providing a novel technological means for the application of UAVs in both military and civilian domains.