融合注意力机制的古建筑青砖损伤检测方法  

Damage detection method in ancient building grey bricks fused with attention mechanisms

作　　者：王茹[1,2] 林浩杰[1] 黄炜 刘奚卓[1] WANG Ru;LIN Haojie;HUANG Wei;LIU Xizhuo(School of Civil Engineering,Xi'an University of Architecture and Technology,Xi'an 710055,China;Key Laboratory of Structural Engineering and Seismic Education,Xi'an 710055,China)

机构地区：[1]西安建筑科技大学土木工程学院,西安710055 [2]结构工程与抗震教育部重点实验室,西安710055

出　　处：《安全与环境学报》2024年第11期4244-4252,共9页Journal of Safety and Environment

基　　金：国家自然科学基金项目(51978566)。

摘　　要：由于自然环境的长期影响,古建筑墙面青砖不可避免地会遭受各种程度的损害,使得青砖墙体存在严重的安全隐患。为有效识别青砖损伤类型,解决传统识别方法成本较高的问题,提出了一种融合注意力机制的古建筑青砖损伤检测算法。基于YOLOv8n算法进行改进:在特征融合网络中增加浅层特征图流入,以解决损伤检测中细小裂缝和损伤纹理等低级特征识别精度低的问题;在特征融合层引入卷积注意力模块(Convolutional Block Attention Module,CBAM),实现通道特征和空间特征的双重关注,进而提高算法对于不同维度特征的融合能力;在特征提取网络部分,通过置换注意力(Shuffle Attention,SA)对C2f模块进行改进,解决由于深度卷积降维操作造成的特征信息丢失问题,从而进一步提高模型的性能。试验结果表明,改进的算法平均精度均值达到55.1%,相比原始YOLOv8n算法提升了7百分点,较YOLOv7-Tiny提升了10.3百分点,较YOLOv9提升了3.5百分点。改进后的算法有效提高了青砖损伤检测性能,为古建筑青砖病害检测提供了有力的技术支持。Due to the long-term influence of the environment,the grey bricks on the walls of ancient buildings inevitably suffered from various degrees of damage,leading to serious safety hazards.To effectively recognize the damage types of grey bricks and solve the problem of the high cost of traditional recognition methods,this paper proposes a damage detection algorithm for grey bricks of ancient buildings based on improved YOLOv8n.Firstly,to solve the problem of low recognition accuracy of low-level features such as small cracks and damage textures in damage detection,we improved the feature fusion network to increase the inflow of shallow feature maps,so that the feature fusion network can fully fuse the low-level features,thereby improving the detection accuracy of the model for small targets.Then,we introduced the Convolutional Block Attention Module(CBAM)into the feature fusion layer to achieve the purpose of paying attention to channel features and spatial features at the same time,thereby improving the feature fusion ability of different dimensions of the model.Next,in the feature extraction network,we improved the C2f module with Shuffle Attention(SA)to solve the problem of feature information loss caused by deep convolution dimension reduction operation and further improved the performance of the model.Finally,we used the grey brick damage recognition dataset to train and evaluate the models and conducted ablation experiments and comparative experiments.Experimental results show that the mean Average Precision(mAP)metric of the improved algorithm reaches 55.1%,which is 7 percentage points higher than that of the original YOLOv8n model,9.5 percentage points higher than that of YOLOv5n,10.3 percentage points higher than that of YOLOv7-Tiny,and 3.5 percentage points higher than that of YOLOv9.The improved algorithm effectively improves the performance of grey brick damage detection and provides strong technical support for the detection of grey brick diseases in ancient buildings.

关 键 词：安全工程 古建筑 青砖 损伤检测 YOLOv8n 

分 类 号：X924[环境科学与工程—安全科学]