基于生成对抗网络的火焰图像生成研究  

Flame pattern generation based on generative adversarial network

作　　者：杨植凯 王腾[2] 卜乐平[2] 欧阳继能 刘超 YANG Zhi-kai;WANG Teng;BU Le-ping;OUYANG Ji-neng;LIU Chao(Unit No.92823,Sanya 572000,China;College of Electrical Engineering,Naval Univ.of Engineering,Wuhan 430033,China;Inst.of Systems Engineering,Academy of Military Sciences,Beijing 100082,China)

机构地区：[1]92823部队,海南三亚572000 [2]海军工程大学电气工程学院,武汉430033 [3]军事科学院系统工程研究院,北京100082

出　　处：《海军工程大学学报》2022年第5期7-12,共6页Journal of Naval University of Engineering

基　　金：国家自然科学基金资助项目(42174051,41971416)。

摘　　要：针对目前火焰图像生成基于编码控制的缺点,提出以火焰的高度和宽度描述火焰形态,并根据火焰的高度和宽度预先生成相应的二值化矩形,再将编码转换为信息量更多的图形作为生成器的条件输入。实验结果表明:以图形形式控制生成火焰可以有效提高生成火焰高度和宽度的可控性,生成火焰的IOU值可达到0.911,高度及宽度的皮尔逊相关系数分别达到ρ_(h)=0.988,ρ_(w)=0.985,均优于编码控制的火焰图形生成。In view of the disadvantages of fire image generation based on coding control, the flame shape was described by the height and width of the flame, and a binarization rectangle was generated in advance according to the height and width of the flame, thus the coding-form conditional variable was converted into a graph with more information as the conditional input of the generator. Experiments show that controlled by graphic, the controllability of generated flame height and width improve effectively. The IOU value of flame can reach 0.911, and the Pearson coefficients between conditional variable and flame height and width can reach up to 0.988 and 0.985 respectively, which are better than the coding control of flame generation.

关 键 词：生成对抗网络 图像生成 火焰图像 火焰特征 

分 类 号：TM921.5[电气工程—电力电子与电力传动]