Moisture Content Recognition Model of Unsaturated Soil Based on Convolutional Neural Networks
-
摘要: 土体含水率是影响细粒土性质的主要因素.土体表层含水率的快速识别是农业和岩土工程中智能监测和智能建造技术发展中的急迫需求.为了克服传统含水率测量或监测方法无法满足土体表层含水率的实时无损监测的局限性，特研发基于图像的含水率智能识别算法.首先在实验室中收集了4种不同类别的土体、在不同含水率下的表面照片，获得了超过1 400张图片的高质量样本库，为机器学习模型构建奠定了数据基础.然后采用经典的卷积神经网络对土体含水率图像数据集进行学习，建立了土体含水率智能识别模型.模型比选结果表明：基于ResNet34架构的土体含水率识别模型效果最佳，在测试集上的含水率预测平均误差约为2%.该模型初步满足了土体表层含水率的实时无损监测需求，能够为农业和岩土工程中智能监测和智能建造技术发展提供重要手段.Abstract: The moisture content of the soil is the main factor affecting the quality of fine-grained soil. Rapid recognition of soil surface moisture content is an urgent need for developing intelligent monitoring and construction technology in agricultural and geotechnical engineering. In order to overcome the limitation that traditional water content measurement or monitoring methods cannot meet the real-time nondestructive monitoring of soil surface moisture content, an intelligent moisture content recognition algorithm based on the image is developed. Firstly, we collected surface photos of 4 different types of soils under different moisture contents in the laboratory and obtained a high-quality sample library of more than 1 400 pictures, which laid a data foundation for machine learning model construction. Then the classical convolutional neural network is used to learn the image dataset of soil moisture content, and the intelligent recognition model of soil moisture content is established. The model comparison results show that the model based on ResNet34 architecture has the best moisture content recognition effect, and the average error of moisture content prediction on the test set is about 2%. This model basically meets the requirement of real-time nondestructive monitoring of soil surface moisture content and can provide an essential means for the development of intelligent monitoring and construction technology in agricultural and geotechnical engineering.
-
图 2 粉质黏土等试样制备(环刀法)
a.液压压实器；b.压实土样（青海粉质粘土）
Fig. 2. Sample preparation for silty clay soil and other soils (cutting ring method)
图 3 标准砂试样制备(“小盒”法)
a.小盒容器；b.标准砂土样
Fig. 3. Sample preparation for sandy soil ("small box" method)
图 8 所用神经网络的结构及其修改
a.简单网络；b. NiN修改；c.VGG16修改；d. ResNet系列修改
Fig. 8. The structure of the neutral networks and their modifications in this paper
图 9 基于CNN的含水率预测模型构建过程
Fig. 9. Construction process of moisture content prediction model based on CNN
图 11 分离数据集与混合数据集训练结果对比
a.延庆粉质黏土；b.兰州粉土；c.青海粉质黏土；d.标准砂
Fig. 11. Comparison of training results between separate and mixed datasets
图 12 最优模型全土样含水率预测结果
Fig. 12. Optimal model prediction results of all types soil moisture content
图 13 去除异常样本前样本量相对下降率变化情况
Fig. 13. Changes in the relative decline rate of sample size before removing abnormal samples
图 15 去除异常样本前后识别效果对比
Fig. 15. Comparison of identification effect before and after removing abnormal samples
表 1 土样基本性质
Table 1. Basic properties of each soil
土壤种类 比重
Gs(g/cm3)最优含水率
wo(%)最大干密度
$ \rho $d(g/cm3)1.延庆粉质黏土 2.73 16.50 1.80 2.兰州粉土 2.71 14.81 1.78 3.青海粉质黏土 2.68 15.50 1.75 4.标准砂 2.65 11.56 1.82 
下载: 导出CSV
表 2 原数据集总结
Table 2. Summary of original dataset
土壤类型 压实度 平行样本数量（个） 照片数量（张） 延庆粉质黏土 1.00 2 156 0.95 1 78 兰州粉土 1.00 2 216 0.90 1 108 0.85 2 216 青海粉质黏土 0.95 1 132 0.90 1 132 0.85 2 264 标准砂 - 4 108
下载: 导出CSV
表 3 分离数据集总结
Table 3. Summary of separate dataset
土壤类型 训练集 测试集 延庆粉质黏土 1 497 375 兰州粉土 3 456 864 青海粉质黏土 3 379 845 标准砂 173 43
下载: 导出CSV
表 5 五种网络架构模型的复杂度与性能对比
Table 5. Complexity and performance comparison of five network architecture models
模型 模型大小
（MB）训练速度
（s/epoch）简单网络 6.49 13.59 NiN 32.33 20.57 VGG16 275.50 99.71 ResNet18 106.39 36.89 ResNet34 178.45 53.65
下载: 导出CSV
表 6 ResNet34各土样识别结果(%)
Table 6. Identification results (%) of soil samples based on ResNet34 model
土样类型 $ {\gamma }_{1} $ $ {\gamma }_{3} $ $ {\gamma }_{5} $ 延庆粉质黏土 32.55 70.83 86.98 兰州粉土 37.38 72.34 90.97 青海粉质黏土 31.37 70.99 88.80 标准砂 39.29 94.64 100
下载: 导出CSV
表 7 异常样本及对比分析
Table 7. Abnormal samples and comparative analysis
土样编号 1 2 3 4 土样类型 延庆粉质黏土 青海粉质黏土 兰州粉土 兰州粉土 异常样本 
预测值：26.55% 预测值：16.23% 预测值：14.68% 预测值：22.35% 真实值：9.05% 真实值：25.95% 真实值：26.20% 真实值：31.85% 对比样本 
预测值：11.79% 预测值：25.39% 预测值：25.18% 预测值：27.76% 真实值：9.05% 真实值：25.95% 真实值：26.20% 真实值：31.85%
下载: 导出CSV
-
Cai, Y., Zheng, W. G., Zhang, X., et al., 2019. Research on Soil Moisture Prediction Model Based on Deep Learning. PLoS One, 14(4): e0214508. https://doi.org/10.1371/journal.pone.0214508 Canziani, A., Paszke, A., Culurciello, E., 2016. An Analysis of Deep Neural Network Models for Practical Applications. ArXiv, abs/1605.07678. https://doi.org/10.48550/arXiv.1605.07678 Chang, D., Li, X., Liu, J. K., et al., 2014. Study Progress and Comparison of Soil Moisture Content Measurement Methods. Geotechnical Investigation & Surveying, 42(9): 17-22, 35 (in Chinese with English abstract). Chen, T. Q., Moreau, T., Jiang, Z. H., et al., 2018. TVM: An Automated End-to-End Optimizing Compiler for Deep Learning. ArXiv, 1802.04799. http://arxiv.org/abs/1802.04799.pdf Cheng, G. J., Li, B., Wan, X. L., et al., 2021. Research on Classification of Rock Section Image Based on Squeezenet Convolutional Neural Network. Mineralogy and Petrology, 41(4): 94-101 (in Chinese with English abstract). Dawson, H. L., Dubrule, O., John, C. M., 2023. Impact of Dataset Size and Convolutional Neural Network Architecture on Transfer Learning for Carbonate Rock Classification. Computers & Geosciences, 171: 105284. https://doi.org/10.1016/j.cageo.2022.105284 Domínguez-Cuesta, M. J., Quintana, L., Valenzuela, P., et al., 2021. Evolution of a Human-Induced Mass Movement under the Influence of Rainfall and Soil Moisture. Landslides, 18(11): 3685-3693. https://doi.org/10.1007/s10346-021-01731-4 Fan, H. Y., Tian, Z. H., Xu, X. B., et al., 2022. Rockfill Material Segmentation and Gradation Calculation Based on Deep Learning. Case Studies in Construction Materials, 17: e01216. https://doi.org/10.1016/j.cscm.2022.e01216 Guo, T., Yu, H. B., 2018. Methods of Determination of Soil Water Content. Inner Mongolia Science Technology & Economy, (3): 66-67 (in Chinese with English abstract). doi: 10.3969/j.issn.1007-6921.2018.03.035 He, K. M., Zhang, X. Y., Ren, S. Q., et al., 2016. Deep Residual Learning for Image Recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, Las Vegas, 770-778. https://doi.org/10.1109/CVPR.2016.90 Hou, X. L., Feng, Y. H., Wu, G. H., et al., 2016. Application Research on Artificial Neural Network Dynamic Prediction Model of Soil Moisture. Water Saving Irrigation, (7): 70-72, 76 (in Chinese with English abstract). doi: 10.3969/j.issn.1007-4929.2016.07.017 Huang, G., Liu, Z., Laurens, V., et al., 2016. Densely Connected Convolutional Networks. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, 2261-2269. https://doi.org/10.1109/CVPR.2017.243 Huang, W. B., Ding, M. T., Wang, D., et al., 2022. Evaluation of Landslide Susceptibility Based on Layer Adaptive Weighted Convolutional Neural Network Model along Sichuan-Tibet Traffic Corridor. Earth Science, 47(6): 2015-2030 (in Chinese with English abstract). Huang, X. H., Li, T. F., Liu, X. X., et al., 2022. Research on Tracking and Prediction of Rock Fissure Sevelopment Based on Improved Faster R-CNN Algorithm. Journal of Henan Polytechnic University (Natural Science), 41(4): 134-141 (in Chinese with English abstract). Krizhevsky, A., Sutskever, I., Hinton, G. E., 2017. ImageNet Classification with Deep Convolutional Neural Networks. Communications of the ACM, 60(6): 84-90. https://doi.org/10.1145/3065386 Lin, M., Chen, Q., Yan, S. C., 2014. Network in Network. ArXiv Preprint ArXiv, 1312.4400. https://doi.org/10.48550/arXiv.1312.4400 McBratney, A. B., Minasny, B., Viscarra Rossel, R., 2006. Spectral Soil Analysis and Inference Systems: A Powerful Combination for Solving the Soil Data Crisis. Geoderma, 136(1/2): 272-278. https://doi.org/10.1016/j.geoderma.2006.03.051 Ni, J. P., Gao, M., Wei, C. F., et al., 2009. Effects of Soil Water Content on Soil Shearing Strength to Different Soil Layer of Shallow Landslide. Journal of Soil and Water Conservation, 23(6): 48-50 (in Chinese with English abstract). doi: 10.3321/j.issn:1009-2242.2009.06.012 Persson, M., 2005. Estimating Surface Soil Moisture from Soil Color Using Image Analysis. Vadose Zone Journal, 4(4): 1119-1122. https://doi.org/10.2136/vzj2005.0023 Simonyan, K., Zisserman, A., 2015. Very Deep Convolutional Networks for Large-Scale Image Recognition. International Conference on Learning Representations (ICLR), San Diago, 1-14. https://doi.org/10.48550/arXiv.1409.1556 Song, B. H., Chen, W. W., Wu, W. J., et al., 2017. Experimental Study on the Dynamic Properties of Sliding Zone Soil of a Landslide under Varying Water Content. China Earthquake Engineering Journal, 39(4): 744-749, 758 (in Chinese with English abstract). doi: 10.3969/j.issn.1000-0844.2017.04.0744 Sun, Y. L., Qu, Z. Y., Liu, Q. M., et al., 2020. Dynamic Study of Regional Soil Moisture Content Based on Multi-Source Remote Sensing Co-Inversion. Journal of Southwest University (Natural Science Edition), 42(12): 46-53 (in Chinese with English abstract). Szegedy, C., Liu, W., Jia, Y., et al., 2015. Going Deeper with Convolutions. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Chicago, 1-9. https://doi.org/10.48550/arXiv.1409.4842 Tian, X., Daigle, H., 2018. Machine-Learning-Based Object Detection in Images for Reservoir Characterization: A Case Study of Fracture Detection in Shales. The Leading Edge, 37(6): 435-442. https://doi.org/10.1190/tle37060435.1 Wang, D. L., Shu, Y. G., 2017. Research Progress in Determination Methods for Soil Water Content. Journal of Mountain Agriculture and Biology, 36(2): 61-65 (in Chinese with English abstract). Wang, M., Li, X., Chen, L. H., et al., 2020. A Modified Soil Water Content Measurement Technique Using Actively Heated Fiber Optic Sensor. Journal of Rock Mechanics and Geotechnical Engineering, 12(3): 608-619. https://doi.org/10.1016/j.jrmge.2019.11.003 Wang, Z. L., Wang, X. Y., Sun, L., 2012. Research of Variation Laws of Shear Strength between Invaded and Natural Collapsible Loess—A Case Study for Collapsible Loess in Gongyi of Henan. South-to-North Water Diversion and Water Science & Technology, 10(3): 123-126 (in Chinese with English abstract). Yao, L., Wang, Y. R., Chen, Z., et al., 2020. Time-Varying Characteristics of Spatial Variability of Soil Moisture Content in Farmland. Water Saving Irrigation, (3): 33-39 (in Chinese with English abstract). Yi, X. F., Li, H., Zhang, R. C., et al., 2022. Rock Mass Structural Surface Trace Extraction Based on Transfer Learning. Open Geosciences, 14(1): 98-110. https://doi.org/10.1515/geo-2022-0337 Yu, D. C., Ding, P., Yang, Z. Q., et al., 2021. Study on the Influence of Initial Water Content on the Breakout of Loess Landslide Barrier Dam. Mountain Research, 39(3): 367-377 (in Chinese with English abstract). Zanetti, S. S., Cecílio, R. A., Alves, E. G., et al., 2015. Estimation of the Moisture Content of Tropical Soils Using Colour Images and Artificial Neural Networks. CATENA, 135: 100-106. https://doi.org/10.1016/j.catena.2015.07.015 Zhang, H., Wei, W. L., Liu, S. S., et al., 2023. Intelligent Identification Method of Moisture Content of Loess Based on Transfer Convolutional Neural Networks. Journal of Engineering Geology, 31(1): 21-31 (in Chinese with English abstract). Zhao, N. Y., Jiang, Y., Song, Y., 2021. Recognition and Classification of Concrete Cracks under Strong Interference Based on Convolutional Neural Network. Traitement Du Signal, 38(3): 911-917. https://doi.org/10.18280/ts.380338 Zhao, Y. F., Wang, C. S., 2017. Principles and Comparison of Common Methods for Determining Soil Water Content. Horticulture & Seed, 37(10): 70-73 (in Chinese with English abstract). Zheng, Z. W., Gao, Z. Z., 2020. Research on Influence of Soil Moisture Content of Farmland on Infiltration Model Parameters. E3S Web of Conferences, 189(3): 01011. https://doi.org/10.1051/e3sconf/202018901011 Zhou, C., Hu, G. L., Deng, L. Y., et al., 2021. Variation Characteristics of Soil Organic Matter and Water Content in Different Types of Landscape in the Middle Reaches of Heihe River. Journal of Gansu Agricultural University, 56(4): 126-135 (in Chinese with English abstract). Zhou, S. G., 2007. Main Geology Disasters Type of Loess Tunnel. Geology and Prospecting, 43(2): 103-107 (in Chinese with English abstract). 常丹, 李旭, 刘建坤, 等, 2014. 土体含水率测量方法研究进展及比较. 工程勘察, 42(9): 17-22, 35. https://www.cnki.com.cn/Article/CJFDTOTAL-GCKC201409004.htm 程国建, 李碧, 万晓龙, 等, 2021. 基于SqueezeNet卷积神经网络的岩石薄片图像分类研究. 矿物岩石, 41(4): 94-101. https://www.cnki.com.cn/Article/CJFDTOTAL-KWYS202104009.htm 郭焘, 于红博, 2018. 土壤含水量测定方法综述. 内蒙古科技与经济, (3): 66-67. https://www.cnki.com.cn/Article/CJFDTOTAL-NMKJ201803035.htm 侯晓丽, 冯跃华, 吴光辉, 等, 2016. 基于人工神经网络土壤墒情动态预测模型应用研究. 节水灌溉, (7): 70-72, 76. https://www.cnki.com.cn/Article/CJFDTOTAL-JSGU201607017.htm 黄武彪, 丁明涛, 王栋, 等, 2022. 基于层数自适应加权卷积神经网络的川藏交通廊道沿线滑坡易发性评价. 地球科学, 47(6): 2015-2030. 黄晓红, 李铁锋, 刘祥鑫, 等, 2022. 基于改进Faster R-CNN算法的岩石裂隙发展方向跟踪预测研究. 河南理工大学学报(自然科学版), 41(4): 134-141. https://www.cnki.com.cn/Article/CJFDTOTAL-JGXB202401009.htm 倪九派, 高明, 魏朝富, 等, 2009. 土壤含水率对浅层滑坡体不同层次土壤抗剪强度的影响. 水土保持学报, 23(6): 48-50. https://www.cnki.com.cn/Article/CJFDTOTAL-TRQS200906014.htm 宋丙辉, 谌文武, 吴玮江, 等, 2017. 不同含水率下滑坡滑带土动力特性试验研究. 地震工程学报, 39(4): 744-749, 758. https://www.cnki.com.cn/Article/CJFDTOTAL-ZBDZ201704024.htm 孙宇乐, 屈忠义, 刘全明, 等, 2020. 基于多源遥感协同反演的区域性土壤含水率动态研究. 西南大学学报(自然科学版), 42(12): 46-53. https://www.cnki.com.cn/Article/CJFDTOTAL-XNND202012005.htm 王大龙, 舒英格, 2017. 土壤含水量测定方法研究进展. 山地农业生物学报, 36(2): 61-65. https://www.cnki.com.cn/Article/CJFDTOTAL-SDNS201702013.htm 王志良, 王秀艳, 孙琳, 2012. 湿陷性黄土侵水前后抗剪强度变化规律研究: 以河南巩义地区湿陷性黄土为例. 南水北调与水利科技, 10(3): 123-126. https://www.cnki.com.cn/Article/CJFDTOTAL-NSBD201203028.htm 姚丽, 王仰仁, 陈钊, 等, 2020. 农田土壤含水率空间变异的时变特性研究. 节水灌溉, (3): 33-39. https://www.cnki.com.cn/Article/CJFDTOTAL-JSGU202003007.htm 雨德聪, 丁攀, 杨志全, 等, 2021. 初始含水率对黄土滑坡堰塞坝溃决影响研究. 山地学报, 39(3): 367-377. https://www.cnki.com.cn/Article/CJFDTOTAL-SDYA202103007.htm 张晗, 魏文龙, 刘森森, 等, 2023. 基于迁移卷积神经网络的黄土含水率智能识别. 工程地质学报, 31(1): 21-31. https://www.cnki.com.cn/Article/CJFDTOTAL-GCDZ202301003.htm 赵宇飞, 王长沙, 2017. 常用土壤含水量测定方法的原理及比较. 园艺与种苗, 37(10): 70-73. https://www.cnki.com.cn/Article/CJFDTOTAL-GNZL201710026.htm 周川, 胡广录, 邓丽媛, 等, 2021. 黑河中游不同景观类型土壤有机质与含水率变化特征. 甘肃农业大学学报, 56(4): 126-135. https://www.cnki.com.cn/Article/CJFDTOTAL-GSND202104017.htm 周尚国, 2007. 黄土隧道的主要地质灾害类型. 地质与勘探, 43(2): 103-107. https://www.cnki.com.cn/Article/CJFDTOTAL-DZKT200702021.htm -
点击查看大图
计量
- 文章访问数: 699
- HTML全文浏览量: 268
- PDF下载量: 76
- 被引次数: 0
