融合深度学习聚类分割和形态学的混凝土表面裂缝量化识别 下载: 1192次
Quantitative Identification of Concrete Surface Cracks Based on Deep Learning Clustering Segmentation and Morphology
1 武汉理工大学安全与应急管理学院, 湖北 武汉 430079
2 重庆市计量质量检测研究院, 重庆 404100
图 & 表
图 1. 裂缝识别网络(CIN)结构示意图
Fig. 1. Schematic of Crack Identification Net (CIN)
下载图片 查看原文
图 2. 识别流程图
Fig. 2. Flowchart of identification
下载图片 查看原文
图 3. 裂缝与非裂缝图像。(a)裂缝图像;(b)非裂缝图像
Fig. 3. Example of crack and non-crack images. (a) Crack images; (b) non-crack images
下载图片 查看原文
图 4. 第4组模型的训练以及验证结果。(a)训练结果;(b)验证结果
Fig. 4. Training and validation results for group 4th model. (a) Training results; (b) validation results
下载图片 查看原文
图 5. 5种不同组合的训练与验证正确率
Fig. 5. Accuracy rate of training and validation for 5 different groups
下载图片 查看原文
图 6. 裂缝识别效果。(a)原始图像;(b)裂缝分类与识别结果
Fig. 6. Example of identification results. (a) Original image; (b) crack classification and identification result
下载图片 查看原文
图 7. 利用本文算法和传统方法获得的分割效果。(a)原图像;(b)改进Otsu算法;(C)改进Canny算法;(d)改进中值滤波算法;(e)本文算法
Fig. 7. Segmentation results obtained by the proposed algorithm and traditional methods. (a) Original image; (b) improved Otsu algorithm; (c) improved Canny algorithm; (d) improved median filter algorithm; (e) our algorithm
下载图片 查看原文
图 8. 各算法的评价指标对比
Fig. 8. Comparison of evaluation indicators of each algorithm
下载图片 查看原文
图 9. 利用本文算法和聚类方法获得的分割效果。(a)原图像;(b) K-means算法;(c) Meansshift算法;(d) Fuzzy C-means算法;(e)本文算法
Fig. 9. Segmentation results obtained by the proposed algorithm and clustering methods. (a) Original image; (b) K-means algorithm; (c) mean shift algorithm; (d) fuzzy C-means algorithm; (e) our algorithm
下载图片 查看原文
图 10. 各算法的评价指标对比
Fig. 10. Comparison of evaluation indicators of each algorithm
下载图片 查看原文
图 11. 裂缝标识示例。(a)原图像;(b)神经网络识别;(c)分割;(d)标识
Fig. 11. Identification of cracks with different thicknesses. (a) Original image; (b) identification of neural network; (c) segmentation; (d) mark
下载图片 查看原文
图 12. 裂缝标识示例图。(a)裂缝1;(b)裂缝2
Fig. 12. Example of crack marking. (a) Crack 1; (b) crack 2
下载图片 查看原文
图 13. 用于量化计算的裂缝原始图像
Fig. 13. Original crack images for quantitative calculation
下载图片 查看原文
表 1裂缝1和2的像素尺寸
Table1. Pixel sizes of crack 1 and crack 2
Crack number | Segmentationnumber | Width /pixel | Length /pixel | Averagewidth /pixel | Overalllength /pixel | Area /pixel2 | Occupationration /% |
---|
| 1 | 4 | 107 | | | | | | 2 | 4 | 63 | | | | | | 3 | 4 | 97 | | | | | Crack 1 | 4 | 5 | 152 | 4.63 | 1193 | 5994 | 0.76 | | 5 | 6 | 163 | | | | | | 6 | 6 | 286 | | | | | | 7 | 5 | 258 | | | | | | 8 | 3 | 67 | | | | | | 1 | 5 | 913 | | | | | Crack 2 | 2 | 6 | 106 | 4.75 | 1384 | 6213 | 1.47 | | 3 | 4 | 240 | | | | | | 4 | 4 | 125 | | | | |
|
查看原文
表 2裂缝1和2的实际尺寸
Table2. Actual size of cracks 1 and crack 2
Crack number | Quantitative calculation | Crack gauge measurement |
---|
Averagewidth /mm | Overalllength /mm | Area /mm2 | Averagewidth /mm | Overalllength /mm | Area /mm2 |
---|
Crack 1 | 0.97 | 250.53 | 264.33 | 1.00 | 251.20 | 261.52 | Crack 2 | 1.00 | 290.64 | 273.99 | 0.98 | 288.42 | 270.26 |
|
查看原文
表 3统计结果对比
Table3. Comparison of statistical results
Group number | Average width /mm | Overall length /mm | Area /mm2 |
---|
Quantitativecalculation | Crack gaugemeasurement | Error | Quantitativecalculation | Crack gaugemeasurement | Error | Quantitativecalculation | Crack gaugemeasurement | Error |
---|
1 | 0.98 | 1.00 | 0.02 | 253.11 | 257.33 | 4.22 | 250.52 | 254.33 | 3.81 | 2 | 0.9 | 0.92 | 0.02 | 232.45 | 236.62 | 4.17 | 211.20 | 215.69 | 4.49 | 3 | 1.12 | 1.08 | 0.04 | 289.27 | 280.94 | 8.33 | 320.98 | 318.41 | 2.57 | 4 | 1.05 | 1.02 | 0.03 | 271.19 | 265.44 | 5.75 | 281.74 | 277.74 | 4 | 5 | 0.97 | 1.00 | 0.03 | 250.53 | 256.28 | 6.75 | 248.82 | 253.28 | 4.46 | 6 | 1.02 | 1.00 | 0.02 | 263.44 | 259.60 | 3.84 | 266.70 | 260.24 | 6.46 | 7 | 0.95 | 0.98 | 0.03 | 245.36 | 250.42 | 5.06 | 240.12 | 247.41 | 7.29 | 8 | 0.95 | 0.97 | 0.02 | 246.72 | 251.53 | 4.81 | 244.54 | 248.78 | 4.24 | 9 | 0.99 | 1.02 | 0.03 | 265.7 | 263.44 | 2.26 | 265.88 | 267.23 | 1.35 | 10 | 1.2 | 1.16 | 0.04 | 309.93 | 304.60 | 5.33 | 364.81 | 355.60 | 9.21 |
|
查看原文
表 4统计结果精度
Table4. Accuracy of statistical results
Group number | Accuracy of average width /mm | Accuracy of overall length /mm | Accuracy of area /mm2 |
---|
1 | 98.00 | 98.36 | 98.50 | 2 | 97.83 | 98.24 | 97.92 | 3 | 96.30 | 97.03 | 99.19 | 4 | 97.06 | 97.83 | 98.56 | 5 | 97.00 | 97.37 | 98.24 | 6 | 98.00 | 98.52 | 97.52 | 7 | 96.94 | 97.98 | 97.05 | 8 | 97.94 | 98.09 | 98.30 | 9 | 97.06 | 99.14 | 99.49 | 10 | 96.55 | 98.25 | 97.41 |
|
查看原文
杨杰文, 章光, 陈西江, 班亚. 融合深度学习聚类分割和形态学的混凝土表面裂缝量化识别[J]. 激光与光电子学进展, 2020, 57(22): 221023. Jiewen Yang, Guang Zhang, Xijiang Chen, Ya Ban. Quantitative Identification of Concrete Surface Cracks Based on Deep Learning Clustering Segmentation and Morphology[J]. Laser & Optoelectronics Progress, 2020, 57(22): 221023.