基于自学习框架的红外场景仿真效果评价

Evaluation for infrared scene simulation based on self-learning framework

针对红外场景仿真效果评估困难的问题,提出了基于自学习框架的评价方法。首先,从仿真图像与实际图像视觉对比的角度提出了面向图像视觉相似的红外场景仿真效果评价指标体系,用于量化评价过程;其次,提出以极限学习机(extreme learning machine,ELM)为核心建立评估模型,建立包括蒙特卡罗样本仿真、ELM评估网络及自更新仿真样本评估模型等3部分在内的自学习框架来生成仿真样本、强化对ELM的训练;最后,针对实际样本数量较少的问题,在此框架基础上提出了包括样本评定、自学习、评估模型测试3个阶段在内的仿真图像相似性评估方法,实现了从样本生成到评估过程的自动化。实验结果表明提出的自学习框架能够显著提高评估模型的正确率,而且训练后的评估模型适用性强,可独立自主进行红外场景仿真效果评估。

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.