作战辅助系统中基于深度学习的投影图像颜色补偿方法  

Deep Learning-based Color Compensation Method for Projected Images in Combat Assistant System

作　　者：张凤 张超[1] 杨华民[1] 王发斌 ZHANG Feng;ZHANG Chao;YANG Huamin;WANG Fabin(College of Computer Science and Technology,Changchun University of Science and Technology,Changchun 130022,Jilin,China;College of Computer,Jilin Normal University,Siping 136000,Jilin,China)

机构地区：[1]长春理工大学计算机科学技术学院,吉林长春130022 [2]吉林师范大学计算机学院,吉林四平136000

出　　处：《兵工学报》2021年第11期2418-2423,共6页Acta Armamentarii

基　　金：国家自然科学基金青年科学基金项目(61702051)。

摘　　要：为了在复杂环境下实现显示系统的快速布置,提高现代战争中野外指挥系统建立的机动性,提出一种基于深度学习的投影图像颜色补偿方法,实现大规模数据在任意复杂颜色平面的可视化显示。该方法利用端到端的深度学习网络,基于一个模型和一个目标函数,规避了深度学习中多模块固有的缺陷;通过增加CompenNet网络层数,增加训练模型中图像特征的提取数量;采用改进的损失函数SSIM+Smooth L1计算图像相似度,增强损失函数的鲁棒性和稳定性,同时加快网络的收敛速度。实验结果显示,改进的CompenNet网络在相同24个数据集上训练迭代1000次后,生成的图像峰值信噪比平均值提高5.54%,结构相似性平均值提高0.14%,均方根误差平均值降低0.14%,人眼主观感知投影效果也表现得更好。A deep learning-based color compensation method for projected images is proposed to realize the visual display of large-scale data in arbitrary complex color plane.The proposed method is aimed to realize the rapid deployment of display system in complex environment and improve the mobility of field command system in modern war.The method is used to avoid the inherent defects of multiple modules in deep learning based on an end-to-end deep learning network,a model and an objective function.The number of image features extracted from the training model is increased by deepening the layers of CompenNet.The improved loss function SSIM+Smooth L1 is used to calculate image similarity,which enhances the robustness and stability of the loss function while speeding up the convergence of the network.The experimental results show that the PSNR and SSIM average values of the images generated by the improved CompenNet after 1000 training iterations on the same 24 data sets are increased by 5.54%and 0.14%,respectively,the RMSE average value is decreased by 0.14%,and the subjective perception projection effect of human eyes is also better.

关 键 词：投影图像 颜色补偿 深度学习 作战辅助系统 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]