摘要
研究从水泥稳定碎石基层入手,采集数据并构建数据集,采用Image Labeler软件进行数据标注和数据增广。设计了基于空洞卷积的DeepLabV3+网络结合MobileNetV2的语义分割模型,并使用分水岭方法进一步改进模型分割结果。结果显示,模型准确率最大值为80.31%,最小值为53.77%,在集料和背景图片上的F值分别为0.85127和0.88127。经模型和分水岭方法处理后,图片的静矩变异系数主要集中在0.03~0.17。研究设计的基于空洞卷积的语义分割模型在沥青路面施工均匀性检测上表现出良好的性能,能有效提升检测效果。
The study started with the cement-stabilized crushed stone base,collected data and constructed a dataset,and used Image Labeler software for data annotation and data augmentation.A semantic segmentation model based on the DeepLabV3+network with dilated convolution combined with MobileNetV2 was designed,and the watershed method was further used to improve the model's segmentation results.The results showed that the maximum accuracy of the model was 80.31%,and the minimum was 53.77%.The F-values of the model on aggregate and background images were 0.85127 and 0.88127,respectively.After processing with the model and the watershed method,the static moment variation coefficient of the images was mainly concentrated in the range of 0.03 to 0.17.The semantic segmentation model based on dilated convolution designed in this study showed good performance in the detection of the construction uniformity of asphalt pavement and could effectively improve the detection effect.
出处
《智能城市》
2025年第2期140-142,共3页
Intelligent City
关键词
空洞卷积
路面
施工
均匀性
水稳层
dilated convolution
pavement
construction
uniformity
water stable layer
