中圖分類號：TP389.1 文獻(xiàn)標(biāo)志碼：A DOI: 10.16157/j.issn.0258-7998.245971 中文引用格式： 桑晨浩，莫路鋒，屠國青. 復(fù)雜景觀圖像的語義多狀態(tài)圖像風(fēng)格遷移[J]. 電子技術(shù)應(yīng)用，2025，51(6)：40-46. 英文引用格式： Sang Chenhao，Mo Lufeng，Tu Guoqing. Multi-state image generation of complex landscapes via semantic category style transfer[J]. Application of Electronic Technique，2025，51(6)：40-46.
Multi-state image generation of complex landscapes via semantic category style transfer
Sang Chenhao，Mo Lufeng，Tu Guoqing
College of Mathematics and Computer Science， Zhejiang Agricultural and Forest University
Abstract： Complex landscape images contain various objects with different characteristics, and traditional style transfer methods are unable to perform local style transfer on different objects within the same image. CycleGAN can achieve style transfer without paired samples through a pseudo-supervised strategy. However, CycleGAN fails to transfer styles between different categories of objects in complex landscape images; moreover, it lacks generalization ability in complex scenes and has high complexity. Therefore, this paper proposes a method for generating complex landscape multi-state images based on semantic categories, namely Semantic Category Style Transfer (SCST), which effectively combines local features for the generation of complex landscape images. Additionally, this paper introduces a context-aware style transfer model called GCycleGAN. Experimental results show that the performance of the proposed GCycleGAN is superior to that of deep learning-based image generation models such as CycleGAN, DualGAN, and Munit.
Key words : landscape；local style transfer；SCST；CycleGAN；Gated-MLP；image generation
