基于迁移学习和卷积神经网络的桥梁图像美学评价  

Aesthetic Evaluation of Bridge Images Based on Transfer Learning and Convolutional Neural Networks

作　　者：叶添照 赵少杰 云季彪 YE Tianzhao;ZHAO Shaojie;YUN Jibiao(School of Civil Engineering,Xiangtan University,Xiangtan 411105,China)

机构地区：[1]湘潭大学土木工程学院,湖南湘潭411105

出　　处：《华侨大学学报(自然科学版)》2025年第2期176-182,共7页Journal of Huaqiao University(Natural Science)

摘　　要：为了在桥梁方案设计中实现桥梁美学智能评价,提出一种基于迁移学习和卷积神经网络的桥梁图像美学自动评价方法。首先,通过冻结部分卷积层和修改丢弃率优化VGG16网络模型;其次,利用迁移学习将已知数据集AVA模型运用到桥梁图像评价上,最终可自动输出对应桥梁美学评分值。结果表明:与人工主观评分相比,文中方法的平均吻合度达到90.2%,该智能评价方法具有较好的准确性和工程实用性。In order to realize intelligent evaluation of bridge schemes in bridge design,an automatic aesthetic evaluation method for bridge images based on transfer learning and convolutional neural networks is proposed.First,the VGG16 network model is optimized by freezing part of convolution layers and modifying the dropout rate.Second,the known data set AVA model is applied to bridge images evaluation by transfer learning,which can automaticlly output the corresponding aesthetic scores.The results show that compared with the manual subjective evaluation,the average coincidence degree of the proposed method is 90.2%,indicating that the intelligent evaluation method has good accuracy and engineering practicability.

关 键 词：桥梁美学 卷积神经网络 迁移学习 美学评价 

分 类 号：TU026[建筑科学]