基于卷积神经网络算法的混凝土桥梁裂缝识别与定位技术

Identifying and Positioning Technologies of Concrete Bridge Crack based on Convolutional Neural Network

无人机技术的进步与高性能计算机的出现,促进了桥梁结构智能化检测的发展。为了实现对无人机获取的大量照片的自动化处理,提出了一种基于计算机视觉的混凝土桥梁裂缝识别与定位技术。以卷积神经网络为核心算法,构建相应的数据库,通过对现有方法的改进,提出一种混凝土桥梁裂缝高效识别的技术。用于检测桥梁裂缝的卷积神经网络架构由3组卷积与池化层、两组Dropout与全连接层组成,算法测试集准确率为93.6%。结合卷积神经网络与滑动窗口算法,搭建相应的数据库与网络架构,提出一种混凝土桥梁裂缝准确定位的技术。结果表明,本文所提出的混凝土桥梁裂缝识别与定位技术,计算效率较高,准确度较好,可以直接应用于识别由无人机拍摄得到的桥梁裂缝照片。此项技术加速了识别速度且具有较高的准确率,为智能化、自动化检测桥梁病害奠定了良好的基础。

The progress of Unmanned Aerial Vehicle(UAV)technology and the emergence of highperformance computers have promoted the development of intelligent detection of bridge structures.In order to automatically process a large number of photos acquired by UAV,a computer vision based crack identification and location technology for concrete bridges is proposed.With convolution neural network as the core algorithm,the corresponding database is constructed.By improving the existing methods,an efficient identification technology of concrete bridge cracks is proposed.The convolution neural network architecture for bridge crack detection consists of three groups of convolution and pooling layers,two groups of Dropout and full connection layers.The accuracy of the algorithm test set is 93.6%.Combining convolution neural network with sliding window algorithm,the corresponding database and network architecture are built,and a technique for accurate crack location of concrete bridges is proposed.The results show that the crack identification and location technology proposed in this paper is efficient and accurate,and can be directly applied to identify the bridge crack photos taken by UAV.This technology accelerates the recognition speed and has a high accuracy,which lays a good foundation for intelligent automatic detection of bridge defects.