Efficient image segmentation method for interface heterogeneous materials of grouted asphalt concrete under small sample conditions
Article Text (Baidu Translation)
-
摘要: 建立了一种物理约束与注意力机制耦合的智能分割框架；提出了自适应腐蚀-膨胀预处理算法，结合多权重边缘融合算子，增强界面的梯度响应；设计了ASPP-SE网络架构，通过通道注意力与空间金字塔池化的协同优化，实现孔隙分布与集料纹理的多尺度特征解耦；建立了边缘置信度与分割掩码的联合修正策略，对初始分割结果进行后处理优化。结果表明：在自建的灌入式沥青混凝土小样本数据集上，该方法的分割结果较目前先进的SegFormer深度学习模型在准确率上提升8.4%，精确率上提升6.6%，召回率上提升6.9%，有效提升了边界分割精度。该方法有效解决了材料相态交界面模糊、集料纹理特征等复杂场景下的分割失效问题，在实际工程应用中表现出较好的可靠性与普适性，可为低数据量、高异质性的工程材料图像分析提供新思路。Abstract: An intelligent segmentation framework coupling physical constraints and attention mechanisms was proposed. An adaptive erosion-dilation preprocessing algorithm was proposed, combining with multi-weight edge fusion operators to enhance gradient responses of interfaces. The atrous spatial pyramid pooling-squeeze and excitation (ASPP-SE) network architecture was designed to achieve multi-scale feature decoupling of pore distribution and aggregate texture through synergistic optimization of SE-Block and ASPP. A joint correction strategy integrating edge confidence and segmentation masks was established to optimize initial segmentation results through post-processing. The results demonstrate that on our self-constructed small-sample dataset of grouted asphalt concrete, the proposed method achieves an improvement of 8.4% in accuracy, 6.6% in precision, and 6.9% increase in recall compared to the SegFormer deep learning model, effectively enhancing boundary segmentation accuracy. This method effectively resolves segmentation failures in complex scenarios including blurred material phase interfaces and aggregate texture features, demonstrating superior reliability and generalizability in practical engineering applications. It also provides new insights for analyzing engineering material images with limited data and high heterogeneity.
-
图 16 消融与对比试验可视化结果
Figure 16. Visualization results of ablation and comparative experiments
图 19 不同样本量数据集训练各模型预测结果数据
Figure 19. Data of prediction results of training models on datasets with different sample number
图 20 不同初始样本量下的预测结果数据对比
Figure 20. Prediction result data comparison under different initial sample number
表 1 模型训练软硬件平台参数
Table 1. Parameters of the software and hardware platform for model training
软硬件平台 详细参数信息 CPU Intel（R）core（R）i9-14900KF CPU@3.20 GHz GPU NVIDIA GeForce RTX 4080 CUDA CUDA 12.1 CuDNN CuDNN 8.9.5 LabelMe LabelMe 3.16.7 Pytorch Pytorch 1.10.2 Python Python 3.6.6 
下载: 导出CSV
表 2 主干网络对比试验结果
Table 2. Comparative experimental results of backbone network
主干网络 参数量/ 106个 计算量/（109次浮点运算·s-1） 推理速度/（帧·s-1） MobileNetV4 3.8 2.1 65 MobileNetV3 4.5 2.4 58 ResNet18 11.2 3.8 42 EfficientNet 5.3 2.6 53
下载: 导出CSV
表 3 模型训练参数
Table 3. Model training parameters
参数 参数详细信息 下采样因子 8 冻结训练轮次 50 冻结阶段批量大小 8 解冻训练轮次 450 解冻阶段批量大小 4 初始学习率 5.0×10-4 最小学习率 5.0×10-6 权重衰减 0
下载: 导出CSV
表 4 正负例概念
Table 4. Concepts of positive and negative examples
真实情况 预测情况 正例 负例 正例 真正例（TP） 假负例（FN） 负例 假正例（FP） 真负例（TN）
下载: 导出CSV
表 5 迁移学习试验模型预测结果数据
Table 5. Model prediction result data from transfer learning experiments
方法 测试集类型 准确率/% 精确率/% 召回率/% Edge Mask ASPP-SE 基础数据集 93.1 93.2 90.1 SegFormer 基础数据集 84.7 86.6 83.2 Edge Mask ASPP-SE 基础数据集+ 扩展测试集 85.1 82.8 81.5 SegFormer 基础数据集+ 扩展测试集 52.4 56.9 58.3
下载: 导出CSV
-
[1] ZHAO W T, YANG Q. Study on the applicability of asphalt concrete skeleton in the semi-flexible pavement[J]. Construction and Building Materials, 2022, 327: 126923. doi: 10.1016/j.conbuildmat.2022.126923 [2] 刘志杨, 董泽蛟, 周涛, 等. 基于材料信息学的沥青混合料性能提升综述及展望[J]. 中国公路学报, 2024, 37(4): 98-120.LIU Zhi-yang, DONG Ze-jiao, ZHOU Tao, et al. Review and prospects of performance enhancement of asphalt mixtures based on material informatics[J]. China Journal of Highway and Transport, 2024, 37(4): 98-120. [3] BAI T, CHENG Y X, LI Y Y, et al. Design optimization and performance evaluation of metakaolin based geopolymer filled porous (semi-flexible) asphalt mixture[J]. Construction and Building Materials, 2023, 408: 133611. doi: 10.1016/j.conbuildmat.2023.133611 [4] 邓成, 黄冲, 洪锦祥, 等. 超早强半柔性路面在市政路口的应用研究[J]. 公路工程, 2016, 41(1): 116-119, 138.DENG Cheng, HUANG Chong, HONG Jin-xiang, et al. The study of super-early-strength semi-flexible pavement used in municipal road[J]. Highway Engineering, 2016, 41(1): 116-119, 138. [5] 蔡星. 灌入式半柔性路面材料力学行为的宏细观研究[D]. 南京: 东南大学, 2021.CAI Xing. Macro-mesoscopic study on mechanical behavior of grouted semi-flexible pavement material[D]. Nanjing: Southeast University, 2021. [6] CHEN Y W, BAI T, ZHAN A Y, et al. A study of fine-scale low-temperature cracking in geopolymer grouted porous asphalt mixtures based on real aggregate profile modeling[J]. Construction and Building Materials, 2024, 445: 137897. doi: 10.1016/j.conbuildmat.2024.137897 [7] 崔喆, 张生瑞. 基于CT图像的集料三维棱角性计算方法[J]. 交通运输工程学报, 2017, 17(5): 39-49. https://transport.chd.edu.cn/article/id/201705004CUI Zhe, ZHANG Sheng-rui. Computational method of 3D aggregate angularity based on CT images[J]. Journal of Traffic and Transportation Engineering, 2017, 17(5): 39-49. https://transport.chd.edu.cn/article/id/201705004 [8] FANG X, WANG C, LI H, et al. Influence of mesoscopic pore characteristics on the splitting-tensile strength of cellular concrete through deep-learning based image segmentation[J]. Construction and Building Materials, 2022, 315: 125335. doi: 10.1016/j.conbuildmat.2021.125335 [9] YU Y X, ZHENG N, QIAO T, et al. Distinguishing between natural and recolored images via lateral chromatic aberration[J]. Journal of Visual Communication and Image Representation, 2021, 80: 103295. doi: 10.1016/j.jvcir.2021.103295 [10] WANG Z Q, XIE J G, GAO L, et al. Study on air void characteristics and hydraulic characteristics of porous asphalt concrete based on image processing technology[J]. Geofluids, 2021, 2021: 9432323. [11] NOCK R, NIELSEN F. Statistical region merging[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2004, 26(11): 1452-1458. doi: 10.1109/TPAMI.2004.110 [12] KASS M, WITKIN A, TERZOPOULOS D. Snakes: Active contour models[J]. International Journal of Computer Vision, 1988, 1(4): 321-331. doi: 10.1007/BF00133570 [13] BOYKOV Y, VEKSLER O, ZABIH R. Fast approximate energy minimization via graph cuts[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2001, 23(11): 1222-1239. doi: 10.1109/34.969114 [14] ZHAO W X, WANG W W, FENG X C, et al. A new variational method for selective segmentation of medical images[J]. Signal Processing, 2022, 190: 108292. doi: 10.1016/j.sigpro.2021.108292 [15] STARCK J L, ELAD M, DONOHO D L. Image decomposition via the combination of sparse representations and a variational approach[J]. IEEE Transactions on Image Processing, 2005, 14(10): 1570-1582. doi: 10.1109/TIP.2005.852206 [16] MINAEE S, WANG Y. An ADMM approach to masked signal decomposition using subspace representation[J]. IEEE Transactions on Image Processing, 2019, 28(7): 3192-3204. doi: 10.1109/TIP.2019.2894966 [17] LI C L, LI Y G, HAN Z, et al. A novel multiphase segmentation method for interpreting the 3D mesoscopic structure of asphalt mixture using CT images[J]. Construction and Building Materials, 2022, 327: 127010. doi: 10.1016/j.conbuildmat.2022.127010 [18] LI X L, LV X C, ZHOU Y H, et al. Homogeneity evaluation of hot in-place recycling asphalt mixture using digital image processing technique[J]. Journal of Cleaner Production, 2020, 258: 120524. doi: 10.1016/j.jclepro.2020.120524 [19] MEJÍA A M, ALZATE M A, REYES-ORTIZ O J. Segmentation of aggregate and asphalt in photographic images of pavements[J]. Engineering Transactions, 2021, 69(1): 19-42. [20] ENRÍQUEZ-LEÓN A J, DE SOUZA T D, ARAGÃO F T S, et al. Determination of the air void content of asphalt concrete mixtures using artificial intelligence techniques to segment micro-CT images[J]. International Journal of Pavement Engineering, 2022, 23(11): 3973-3982. doi: 10.1080/10298436.2021.1931197 [21] LIU Z J, HUANG T, LIU G Q. Applicability of a new method for mesoscopic structure segmentation of asphalt mixture based on two-dimensional image[J]. Construction and Building Materials, 2024, 421: 135738. doi: 10.1016/j.conbuildmat.2024.135738 [22] WEI J J, LI H B, WAN C. X-ray CT image segmentation of asphalt concrete based on fuzzy C means[J/OL]. Applied Mechanics and Materials, 2012, 170/171/172/173: 3444-3448. [23] HUYAN J, MA T, LI W, et al. Pixelwise asphalt concrete pavement crack detection via deep learning-based semantic segmentation method[J]. Structural Control and Health Monitoring, 2022, 29(8): e2974. [24] HAN S S, CHEN J, JIANG Y H. Experimental evaluation of convolutional neural networks in asphalt concrete computed tomography scan image analysis[C]//ASCE. Construction Research Congress 2024. Reston: ASCE, 2024: 506-515. [25] PENG Y, YANG H D. Aggregate boundary recognition of asphalt mixture CT images based on convolutional neural networks[J]. Road Materials and Pavement Design, 2024, 25(5): 1127-1143. doi: 10.1080/14680629.2023.2233630 [26] DONG J X, LIU J H, WANG N N, et al. Intelligent segmentation and measurement model for asphalt road cracks based on modified mask R-CNN algorithm[J]. Computer Modeling in Engineering Sciences, 2021, 128(2): 541-564. doi: 10.32604/cmes.2021.015875 [27] LIU Y Y, BAI X T, WANG J F, et al. Image semantic segmentation approach based on DeepLabV3 plus network with an attention mechanism[J]. Engineering Applications of Artificial Intelligence, 2024, 127: 107260. doi: 10.1016/j.engappai.2023.107260 [28] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//IEEE. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2018: 7132-7141. [29] CHEN Y T, TAO J J, LIU L W, et al. Research of improving semantic image segmentation based on a feature fusion model[J]. Journal of Ambient Intelligence and Humanized Computing, 2022, 13(11): 5033-5045. doi: 10.1007/s12652-020-02066-z [30] WANG Z H, WU L. Theoretical analysis of the inductive biases in deep convolutional networks[C]//Curran Associates. 37th Conference on Neural Information Processing Systems, NeurIPS 2023. New Orleans: Curran Associates, 2022: 74289-74338. [31] HE K M, GKIOXARI G, DOLLÁR P, et al. Mask R-CNN[C]//IEEE. 2017 IEEE International Conference on Computer Vision (ICCV). New York: IEEE, 2017: 2980-2988. [32] ARCHANA R, ELIAHIM JEEVARAJ P S. Deep learning models for digital image processing: A review[J]. Artificial Intelligence Review, 2024, 57(1): 11. doi: 10.1007/s10462-023-10631-z [33] LI P P, LIU Z P. GeoBind: Segmentation of nucleic acid binding interface on protein surface with geometric deep learning[J]. Nucleic Acids Research, 2023, 51(10): e60. doi: 10.1093/nar/gkad288 [34] XU X M, TU W P, YANG Y H. CASE-Net: Integrating local and non-local attention operations for speech enhancement[J]. Speech Communication, 2023, 148: 31-39. doi: 10.1016/j.specom.2023.02.006 [35] PARMAR N, VASWANI A, USZKOREIT J, et al. Image transformer[C]//Microtome Publishing. International Conference on Machine Learning. New York: PMLR, 2018: 4055-4064. [36] ZHAO H S, JIA J Y, KOLTUN V. Exploring self-attention for image recognition[C]//IEEE. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2020: 10073-10082. [37] SHORTEN C, KHOSHGOFTAAR T M. A survey on image data augmentation for deep learning[J]. Journal of Big Data, 2019, 6(1): 60. doi: 10.1186/s40537-019-0197-0 [38] WANG C H, HUANG K Y, YAO Y, et al. Lightweight deep learning: an overview[J]. IEEE Consumer Electronics Magazine, 2024, 13(4): 51-64. doi: 10.1109/MCE.2022.3181759 [39] 阳恩慧, 陈强, 李杰, 等. 沥青路面表面纹理重构与构造深度预测模型[J]. 中国公路学报, 2023, 36(6): 14-23.YANG En-hui, CHEN Qiang, LI Jie, et al. Surface texture reconstruction and mean texture depth prediction model of asphalt pavement[J]. China Journal of Highway and Transport, 2023, 36(6): 14-23. [40] 高涛, 邢可, 刘占文, 等. 基于金字塔多尺度融合的交通标志检测算法[J]. 交通运输工程学报, 2022, 22(3): 210-224. doi: 10.19818/j.cnki.1671-1637.2022.03.017GAO Tao, XING Ke, LIU Zhan-wen, et al. Traffic sign detection algorithm based on pyramid multi-scale fusion[J]. Journal of Traffic and Transportation Engineering, 2022, 22(3): 210-224. doi: 10.19818/j.cnki.1671-1637.2022.03.017 [41] KAJKAMHAENG S, CHANTRAPORNCHAI C. SE-SqueezeNet: SqueezeNet extension with squeeze and-excitation block[J]. International Journal of Computational Science and Engineering, 2021, 24(2): 185. doi: 10.1504/IJCSE.2021.115105 [42] CHENG B W, MISRA I, SCHWING A G, et al. Masked-attention mask transformer for universal image segmentation[C]//IEEE. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2022: 1280-1289. [43] LI Z, WANG M, MEI J, et al. Mail: A unified mask-image-language trimodal network for referring image segmentation[J/OL]. 2021, https://doi.org/10.48550/arXiv.2111.10747 .[44] 于华南, 姚丁, 钱国平, 等. 基于细观结构特征的沥青混合料性能数字孪生模型研究综述[J]. 中国公路学报, 2023, 36(3): 20-44.YU Hua-nan, YAO Ding, QIAN Guo-ping, et al. Review of digital twin model of asphalt mixture performance based on mesostructure characteristics[J]. China Journal of Highway and Transport, 2023, 36(3): 20-44. [45] QIN D F, LEICHNER C, DELAKIS M, et al. MobileNetV4: Universal models for the mobile ecosystem[C]//Springer. Computer Vision-ECCV 2024. Berlin: Springer, 2024: 78-96. -
图(22) / 表(5)
计量
- 文章访问数: 21
- HTML全文浏览量: 18
- PDF下载量: 8
- 被引次数: 0
