A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Mixed normal form semantic segmentation pavement crack identification method based on parallel architecture
The invention discloses a mixed normal form semantic segmentation pavement crack identification method based on a parallel architecture, and the method comprises the steps: obtaining the historical data of a pavement inspection image, carrying out the preprocessing of the historical data of the pavement inspection image, and obtaining a large-size image block with unified pixel values; performing crack labeling on the large-size image blocks, and distributing crack category labels for the large-size image blocks to obtain a training data set; a parallel framework is adopted, an attention mechanism neural network branch and a convolutional neural network branch are combined, the training data set is used for road surface crack image training, a road surface crack recognition model is constructed, and the output of the road surface crack recognition model is a road surface crack feature enhancement graph; and inputting the pavement crack feature enhancement graph into a full-connection conditional random field for post-processing to obtain an optimized pavement crack prediction result. According to the invention, pavement crack pixel-level prediction with efficiency and precision considered in real time can be realized.
本发明公开了一种基于平行架构的混合范式语义分割路面裂缝识别方法，包括：获取路面巡检图像历史数据，对所述路面巡检图像历史数据进行预处理，得到像素值统一的大尺寸图像块；对所述大尺寸图像块进行裂缝标注，为所述大尺寸图像块分配裂缝类别标签，得到训练数据集；采用平行架构，结合注意力机制神经网络分支和卷积神经网络分支，使用所述训练数据集进行路面裂缝图像训练，构建路面裂缝识别模型，所述路面裂缝识别模型的输出为路面裂缝特征加强图；将所述路面裂缝特征加强图输入到全连接条件随机场进行后处理，得到优化后的路面裂缝预测结果。本发明能够实现实时兼顾效率和精度的路面裂缝像素级预测。
Mixed normal form semantic segmentation pavement crack identification method based on parallel architecture
The invention discloses a mixed normal form semantic segmentation pavement crack identification method based on a parallel architecture, and the method comprises the steps: obtaining the historical data of a pavement inspection image, carrying out the preprocessing of the historical data of the pavement inspection image, and obtaining a large-size image block with unified pixel values; performing crack labeling on the large-size image blocks, and distributing crack category labels for the large-size image blocks to obtain a training data set; a parallel framework is adopted, an attention mechanism neural network branch and a convolutional neural network branch are combined, the training data set is used for road surface crack image training, a road surface crack recognition model is constructed, and the output of the road surface crack recognition model is a road surface crack feature enhancement graph; and inputting the pavement crack feature enhancement graph into a full-connection conditional random field for post-processing to obtain an optimized pavement crack prediction result. According to the invention, pavement crack pixel-level prediction with efficiency and precision considered in real time can be realized.
本发明公开了一种基于平行架构的混合范式语义分割路面裂缝识别方法，包括：获取路面巡检图像历史数据，对所述路面巡检图像历史数据进行预处理，得到像素值统一的大尺寸图像块；对所述大尺寸图像块进行裂缝标注，为所述大尺寸图像块分配裂缝类别标签，得到训练数据集；采用平行架构，结合注意力机制神经网络分支和卷积神经网络分支，使用所述训练数据集进行路面裂缝图像训练，构建路面裂缝识别模型，所述路面裂缝识别模型的输出为路面裂缝特征加强图；将所述路面裂缝特征加强图输入到全连接条件随机场进行后处理，得到优化后的路面裂缝预测结果。本发明能够实现实时兼顾效率和精度的路面裂缝像素级预测。
Mixed normal form semantic segmentation pavement crack identification method based on parallel architecture
一种基于平行架构的混合范式语义分割路面裂缝识别方法
GUO FENG (author) / LIU JIAN (author) / CHANG HONGLEI (author)
2024-04-09
Patent
Electronic Resource
Chinese
| Wiley | 2022
A complex scene pavement crack semantic segmentation method based on dual-stream framework
| Taylor & Francis Verlag | 2023
Asymmetric dual-decoder-U-Net for pavement crack semantic segmentation
| Elsevier | 2023
Semi-supervised semantic segmentation using cross-consistency training for pavement crack detection
| Taylor & Francis Verlag | 2024