基于卷积神经网络的园林景观设计图着色研究  

Study on Coloring of Landscape Design Drawing Based on Convolutional Neural Network

作　　者：黄海钤 王联国[1] 

机构地区：[1]甘肃农业大学信息科学技术学院,甘肃 兰州

出　　处：《软件工程与应用》2023年第2期221-229,共9页Software Engineering and Applications

摘　　要：针对园林景观设计图人工着色工作量大、效率不高等问题,采用卷积神经网络、迁移学习和图像风格迁移技术建立了一种基于卷积神经网络的园林景观设计图着色模型。该模型采用VGG网络提取园林景观设计图、色彩风格图和生成图的特征,通过最小化总损失函数,实现园林景观设计图的着色和景观季节的色彩变换。实验结果表明,基于VGG-16网络的着色模型的着色效果较好,基于VGG-19网络的着色模型的景观季节色彩变换效果更加优良,基于VGG-16网络的着色模型对园林景观设计图着色和景观季节色彩变换的平均损失分别为0.015和0.011,基于VGG-19网络的着色模型对园林景观设计图着色和景观季节色彩变换的平均损失分别为0.027和0.007,验证了该模型的有效性和实用性,为园林景观设计图着色提供了新方法。Aiming at the problems of heavy workload and low efficiency of manual coloring of landscape design drawings, a landscape coloring model based on convolutional neural network, transfer learning and image style transfer was established. The model used VGG network to extract the features of landscape design map, color style map and generated map, and realized the coloring of landscape design map and the color transformation of landscape seasons by minimizing the total loss function. The experimental results show that the coloring model based on VGG-16 network has better coloring effect, and the coloring model based on VGG-19 network has better effect of landscape seasonal color transformation. The average loss of coloring model based on VGG-16 network to landscape design map coloring and landscape seasonal color transformation is 0.015 and 0.011, respectively. The average loss of the coloring model based on VGG-19 network to the coloring of landscape design drawings and landscape seasonal color transformation is 0.027 and 0.007, respectively, which verifies the validity and practicability of the model and provides a new method for the coloring of landscape design drawings.

关 键 词：图像风格迁移 园林景观设计 卷积神经网络 VGG 着色 季节变换 

分 类 号：TU9[建筑科学]