Page 130 - 水利学报2021年第52卷第3期

Neural Information Processing Systems，2014，3：2672-2680 .
［ 24］ 冈萨雷斯 . 数字图像处理［M］. 3 版 . 阮秋琦，译 . 北京：电子工业出版社，2017 .
［ 25］ OTSU N . Threshold Selection Method from Gray-Level Histograms［J］. IEEE Transactions on Systems Man and
Cybernetics，1979，9（1）：62-66 .
［ 26］ HINTON G，SALAKHUTDINOV R R . Reducing the dimensionality of data with neural networks［J］. Science，
2006，313（5786）：504-507 .
［ 27］ KRIZHEVSKY A，SUTSKEVER I，HINTON G . ImageNet classification with deep convolutional neural networks
［J］. Advances in neural Information Processing Systems，2012，25（2）：1097-1105 .
［ 28］ LECUN Y，BOTTOU L . Gradient-based learning applied to document recognition［J］. Proceedings of the IEEE，
1998，86（11）：2278-2324 .
［ 29］ LECUN Y，BOSER B，DENKER J，et al . Backpropagation applied to handwritten zip code recognition［J］. Neu⁃
ral Computation，1989，1（4）：541-551 .
［ 30］ GU J，WANG Z，KUEN J，et al . Recent advances in convolutional neural networks［J］. Pattern Recognition，
2018，77（1）：354-377 .
［ 31］ 张延亿，汪小刚，邓刚，等 . 级配缩尺对堆石压缩特性影响试验研究［J］. 水利水电技术，2017，48（7）：
116-122 .
［ 32］ 万其微，刘斌云，张延亿，等 . 粗粒料级配缩尺中相似比例对密度影响的试验研究［J］. 水利水电技术，
2020，51（6）：157-163 .
［ 33］ FITZGIBBON A，PILU M，FISHER R B . Direct least square fitting of ellipses［J］. IEEE Trans .patt .anal .mach .
intell，1999，21（5）：476 - 480 .
［ 34］ 张延亿，邓刚，张茵琪，等 . 土石混合料固结湿化变形试验研究［J］. 水利学报，2020，51（11）：1393-1400 .
［ 35］ 沙爱民，王超凡，孙朝云 . 一种基于图像的沥青混合料矿料级配检测方法［J］. 长安大学学报（自然科学
版），2010，30（5）：1-5 .
Intelligent detection of gradation for earth-rockfill materials base on
Deep Otsu Convolutional Neural Network

1 1，2 2 2 2 1
LEI Yumeng ，CHEN Zuyu ，YU Shu ，WEN Yanfeng ，WANG Yujie ，LI Yanlong
（1. State Key Laboratory of Eco-hydraulics in Northwest Arid Region of China，Xi'an University of Technology，Xi'an 710048，China；
2. Department of Geotechnical Engineering，China Institute of Water Resources and Hydropower Research，Beijing 100048，China）

Abstract：The mechanical properties and impermeability of rockfill dam can be directly affected by rational⁃
ity of earth-rockfill materials gradation. At present，he gradation testing mainly relies on artificial screening，
t
which cannot achieve large-scale rapid detecting. Traditional image recognition algorithms do not meet the
accuracy requirements of earth-rockfill materials gradation detecting. Image recognition based on deep learn⁃
ing require massive manual labeled samples， which is difficult for engineering. In this paper，c ombining
with the edge detection algorithm based on Otsu in traditional image recognition and the deep learning
model of convolutional neural network， a Deep Otsu Convolutional Neural Network model is established to
realize rapid detection of gradation by integrating the image recognition of earth-rockfill materials and intelli⁃
t
gent analysis of the gradation data. Using limestone as a typical sample，he gradation data and images are
obtained through eighteen groups of standard screening trials for model training and verification. The results
show that the DO-CNN model can greatly improve the stability and accuracy of gradation detecting com⁃
pared to using the edge detection model based on Otsu algorithm，and realizes the rapid detection of
earth-rockfill materials gradation based on image. For the small earth-rockfill materials under 5mm，he mod⁃
t
el also maintained high accuracy.
Keywords： gradation of earth-rockfill materials； DO-CNN； artificial screening trials； small touching parti⁃
i
cles；mage recognition；rockfill dam
（责任编辑：耿庆斋）

— 380 —