Abstract: In order to reduce the influence of subjective cognition on infrared camouflage simulation evaluation, an evaluation method based on selflearning framework is proposed. Firstly, from the perspective of visual comparation between the simulation image and its corresponding actual image to measure the simulation results, focusing on visual similarity, the infrared camouflage simulation evaluation index system is established. Secondly, an evaluation model is established with extreme learning machine (ELM), and a selflearning framework is proposed to generate simulation samples and strengthen the training of ELM, including MonteCarlo simulation, ELM evaluation network and a selfupdating evaluation model for simulation samples, Thirdly, due to the small number of actual samples, a new method of evaluating the similarity of simulation images, including three stages of the sample evaluation, selflearning and network evaluation is puts forward on the basis of the artificial evaluation experiment. Realizing the automation from samples generation to evaluation. Experimental results not only confirm that the proposed selflearning framework can significantly improve the accuracy of the evaluation model, but also show that the practiced evaluation model can be used to evaluate the simulation effects of infrared camouflage independently with strong applicability.