基于特征调节的深度学习模型在桥梁病害识别中的应用

侯怡; 钱松荣; 李雪梅

基于特征调节的深度学习模型在桥梁病害识别中的应用

投稿时间：2024-01-23 修订日期：2024-04-04 点此下载全文

引用本文：

摘要点击次数: 21

全文下载次数: 0

作者	单位	邮编
侯怡^*	贵州大学机械工程学院	550025
钱松荣	贵州大学公共大数据国家重点实验室
李雪梅	贵州大学机械工程学院

基金项目:贵州省科技计划资助项目(黔科合平台人才[2019]5802号)

中文摘要:针对桥梁病害识别率较低的问题，本文提出了一种基于特征调节的病害识别网络（DRNet）以实现对桥梁病害的精准识别，该网络通过改进的全局注意力模块来捕获病害图像的长距离依赖关系并聚集局部关键区域特征，为了获得更加丰富的表征信息，本文还提出了一个特征融合模块用于提取并集成桥梁病害的多尺度特征。同时本文基于实际拍摄构建了一个桥梁病害数据集，并进行了相应的实验分析。实验结果表明，本文的方法给模型的识别性能带来的平均总增益为6.43%，DRNet在三类桥梁病害上的识别准确率均得到了不同程度的提升，且模型的最高识别准确率达到了95.93%。

中文关键词:病害识别注意力机制特征交互特征融合

Deep Learning Models Based on Feature Conditioning in Bridge Disease Identification

Abstract:To address the problem of low recognition rate of bridge diseases, this paper proposes a disease recognition network (DRNet) based on feature conditioning to achieve accurate recognition of bridge diseases, which captures the long-range dependencies of disease images and aggregates local key-area features through an improved global attention module, and in order to obtain richer characterisation information, this paper also proposes a feature fusion module for extracting and integrate the multi-scale features of bridge diseases. Meanwhile, this paper constructs a bridge disease dataset based on actual shooting and carries out the corresponding experimental analysis. The experimental results show that this paper's method brings an average total gain of 6.43% to the recognition performance of the model, and the recognition accuracies of DRNet on the three types of bridge diseases are improved to different degrees, and the highest recognition accuracy of the model reaches 95.93%.

keywords:disease identification attention mechanisms feature interaction feature fusion

查看全文 查看/发表评论 下载pdf阅读器