conference paper
Sketched Irregular Military Symbols Recognition via Generative Adversarial Nets
Abstract
Military situation perception is the key component to accommodate information warfare and enhance command and decision capabilities. Sketched irregular military symbols can swiftly share battlefield information and comprehensively display the combat situation, which is one of the important tasks of the situation perception system. However, the weakness of sketched irregular military symbols recognition is that the major approach is still based on database correlation technology and computer graphics, which result in the inability to accurately identify some situation, leading to a deficit in information exchange. In response to the aforementioned challenges, this paper introduces a novel methodology leveraging Generative Adversarial Networks (GANs). The approach involves cyclic adversarial generation of sketched symbols, which enhances the robustness of the module for sketched irregular military symbols recognition. We design a specific network architecture called RSM-GAN, which including ResNet-based encoder, Style-based generator and Markovianity-based discriminator. In addition, this paper assembles and constructs a new dataset of sketched irregular military symbols, named MMS-Sketch. The experimental results demonstrate that accurate recognition can be achieved on the dataset using the proposed framework. © 2023 IEEE.
Author keywords
composite generative model, generative adversarial network, military situation perception, sketched symbols, style transfer