Algorithm to detect thin strip-shaped structural diseases on airport pavement in complex background
-
摘要:
机场道面裂缝、角隅断裂、接缝破碎、修补等病害宽度狭小、长短不一、图像中像素占比少,呈细带状结构,且与复杂背景对比度低,现有检测算法效果不佳。针对以上问题,提出了一种基于注意力机制与特征融合的深度神经网络模型DetMSPNet。首先,利用注意力机制模块CBAM,使得特征学习更加专注于细带状结构病害区域,抑制干扰信息;其次,构建残差空洞金字塔模块,提取不同尺度空间下的特征信息;然后,设计最大池化支路,便于之后浅、深层不同层次特征进行融合,加强模型对于病害的定位能力,并且将深层特征输入3种不同扩张率的扩张卷积和金字塔池化模块,使得病害特征包含更多全局上下文信息;最后,对所有层输出的病害特征信息进行融合,实现不同尺度、不同层次特征的信息互补。与目前3种经典的目标检测算法在机场道面病害图像数据集APD上做了对比实验,结果表明:所提算法的mAP达到78.51%,优于对比算法。所提DetMSPNet模型,提高了算法对机场道面细带状结构病害检测中宽度狭小、长短不一、图像中像素占比少、与复杂背景对比度低等情况的适应能力。
-
关键词:
- 机场道面细带状结构病害 /
- DetMSPNet /
- 注意力机制 /
- 特征融合 /
- 复杂背景
Abstract:The structural diseases on airport pavement, such as cracks, corner fractures, broken seams, and repairs, have the characteristics of narrow width, different length, and less pixel proportion in the image, which show a thin strip-shaped structure, and the contrast is low in the complex background. These factors lead to detection failure when using the existing detection algorithms. To solve these problems, a deep neural network model, named as DetMSPNet, based on attention mechanism and feature fusion is proposed. First, the attention mechanism module CBAM is used to make the feature learning more focused on the disease area of the thin strip-shaped structure and suppress the interference information. Second, the residual atrous pyramid module is constructed to extract the feature information with different scales. Then, the maximum pooling branch is designed to facilitate the fusion of different features from shallow and deep layers, and improve the positioning ability of the model for diseases. In addition, the deep features are fed into three dilated convolutions with different dilated rates and pyramid pooling modules, so that the disease features contain more global context information. Finally, the disease features generated from all levels are fused to fulfil the information complementation from different scales and different levels. The comparative experiment was conducted with three classical object detection algorithms on APD dataset, and the results show that the proposed algorithm achieves an mAP of 78.51%, which is better than its counterparts. The proposed DetMSPNet improves the adaptability of the algorithm to the narrow width, different length, less proportion of pixels in the image and low contrast with complex background for the detection of thin strip-shaped structural diseases on airport pavement. The experimental results show that the average detection accuracy of the proposed algorithm is improved.
-
表 1 APD数据集详情
Table 1. Details of APD dataset
参数 训练 验证 测试 图像大小/像素 900×600 900×600 900×600 图像数量(扩充后)/张 7 213 1 803 547 表 2 机场道面APD数据集检测结果
Table 2. Detection results on airport pavement APD dataset
算法 网络 平均速度/s AP/% mAP/% 裂缝 接缝破碎 角隅断裂 修补 STDN Densenet169 0.023/(GPU) 28.96 59.42 89.37 79.47 64.28 Faster R-CNN ResNet-101 0.062(GPU) 40.86 68.53 90.47 74.34 68.55 Cascade R-CNN DetNet 0.137(GPU) 47.96 87.11 90.91 82.95 77.23 本文算法 DetMSPNet 0.167(GPU) 49.44 87.54 92.64 84.42 78.51 表 3 CBAM模块的有效性
Table 3. Effectiveness of CBAM module
网络 AP/% mAP/% 裂缝 接缝破碎 角隅断裂 修补 DetMSPNet (无CBAM) 49.82 89.09 90.91 83.67 78.37 DetMSPNet 49.44 87.54 92.64 84.42 78.51 表 4 DetASPP模块的有效性
Table 4. Effectiveness of DetASPP module
网络 AP/% mAP/% 裂缝 接缝破碎 角隅断裂 修补 DetMSPNet (无DetASPP) 48.55 88.35 92.81 83.19 78.23 DetMSPNet 49.44 87.54 92.64 84.42 78.51 表 5 最大池化支路的有效性
Table 5. Effectiveness of maximum pooling branch
网络 AP/% mAP/% 裂缝 接缝破碎 角隅断裂 修补 DetMSPNet(无最大池化支路) 49.18 87.81 90.91 83.19 77.77 DetMSPNet 49.44 87.54 92.64 84.42 78.51 表 6 三种不同扩张率的扩张卷积的有效性
Table 6. Effectiveness of dilated convolution with three dilated rates
网络 AP/% mAP/% 裂缝 接缝破碎 角隅断裂 修补 DetMSPNet(无扩张卷积) 48.75 88.31 92.01 83.94 78.25 DetMSPNet 49.44 87.54 92.64 84.42 78.51 表 7 金字塔池化模块的有效性
Table 7. Effectiveness of pyramid pooling module
网络 AP/% mAP/% 裂缝 接缝破碎 角隅断裂 修补 DetMSPNet(无PPM) 48.06 88.69 90.91 83.32 77.75 DetMSPNet 49.44 87.54 92.64 84.42 78.51 -
[1] SORNCHAREAN S, PHIPHOBMONGKOL S. Crack detection on asphalt surface image using enhanced grid cell analysis[C]//4th IEEE International Symposium on Electronic Design, Test and Applications. Piscataway: IEEE Press, 2008: 49-54. [2] 李清泉, 邹勤, 毛庆洲. 基于最小代价路径搜索的路面裂缝检测[J]. 中国公路学报, 2010, 23(6): 28-33. doi: 10.3969/j.issn.1001-7372.2010.06.005LI Q Q, ZOU Q, MAO Q Z. Pavement crack detection based on minimum cost path searching[J]. China Journal of Highway and Transport, 2010, 23(6): 28-33(in Chinese). doi: 10.3969/j.issn.1001-7372.2010.06.005 [3] GUAN H, LI J, YU Y, et al. Iterative tensor voting for pavement crack extraction using mobile laser scanning data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2014, 53(3): 1527-1537. [4] PENG L, CHAO W, SHUANGMIAO L, et al. Research on crack detection method of airport runway based on twice-threshold segmentation[C]//2015 Fifth International Conference on Instrumentation and Measurement, Computer, Communication and Control (IMCCC). Piscataway: IEEE Press, 2015: 1716-1720. [5] SIMONYAN K, ZISSERMAN A. Two-stream convolutional networks for action recognition in videos[C]//Advances in Neural Information Processing Systems, 2014: 568-576. [6] ZEILER M D, FERGUS R. Visualizing and understanding convolutional networks[C]//European Conference on Computer Vision. Berlin: Springer, 2014: 818-833. [7] GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2014: 580-587. [8] GIRSHICK R. Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2015: 1440-1448. [9] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. doi: 10.1109/TPAMI.2016.2577031 [10] ZHOU P, NI B, GENG C, et al. Scale-transferrable object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2018: 528-537. [11] CAI Z, VASCONCELOS N. Cascade R-CNN: Delving into high quality object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2018: 6154-6162. [12] LI Z, PENG C, YU G, et al. DetNet: A backbone network for object detection[EB/OL]. (2018-04-17)[2020-09-01]. https://arxiv.org/abs/1804.06215v2. [13] 中国民用航空局. 民用机场道面评价管理技术规范: MH/T 5024—2009[S]. 北京: 人民交通出版社, 2009.Civil Aviation Administration of PRC. Technical specifications of aerodrome pavement evaluation and management: MH/T 5024—2009[S]. Beijing: China Communications Press, 2009(in Chinese). [14] 王森, 伍星, 张印辉, 等. 基于深度学习的全卷积网络图像裂纹检测[J]. 计算机辅助设计与图形学学报, 2018, 30(5): 859-867. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJF201805013.htmWANG S, WU X, ZHANG Y H, et al. Image crack detection with fully convolutional network based on deep learning[J]. Journal of Computer-Aided Design & Computer Graphics, 2018, 30(5): 859-867(in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-JSJF201805013.htm [15] CHA Y J, CHOI W, BUYUKOZTURK O. Deep learning-based crack damage detection using convolutional neural networks[J]. Computer-Aided Civil and Infrastructure Engineering, 2017, 32(5): 361-378. doi: 10.1111/mice.12263 [16] CHA Y J, CHOI W, SUH G, et al. Autonomous structural visual inspection using region-based deep learning for detecting multiple damage types[J]. Computer-Aided Civil and Infrastructure Engineering, 2018, 33(9): 731-747. doi: 10.1111/mice.12334 [17] ZOU Q, ZHANG Z, LI Q, et al. DeepCrack: Learning hierarchical convolutional features for crack detection[J]. IEEE Transactions on Image Processing, 2018, 28(3): 1498-1512. [18] WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional block attention module[C]//European Conference on Computer Vision. Berlin: Springer, 2018: 3-19. [19] ZHAO H, SHI J, QI X, et al. Pyramid scene parsing network[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2017: 2881-2890.