-
摘要:
以常见的16类矿物作为研究对象，收集每一类矿物约1000张图像，按比例划分为训练集、验证集和测试集，通过图像随机选取增加数据的多样性，建立矿物识别InceptionV3模型，训练7万次在测试集上获得81%的识别正确率。通过对损失函数的改进，引入Center Loss损失函数，训练40万次识别准确率提高到86%。对分类的混淆矩阵分析发现，孔雀石等外观特征明显的矿物识别正确率很高，而闪锌矿等与其他矿物容易混淆导致正确率较低。从特征图分析看出，模型很好地提取了孔雀石的放射状特征，矿物图像特征向量聚集程度很高，也说明了模型的可靠性。
Abstract:To study 16 kinds of common minerals, the authors collected 1000 images for each type, and then divided them into training set, validation set and test set. Before putting the images into the model, the authors selected a random area of each image for data augmentation. After training the InceptionV3 model with 70000 steps, the authors obtained an 81% accuracy in the test set. Through improving the loss function and introducing the Center Loss, the authors raised the accuracy to 86% after training 400000 steps. The obfuscation matrix shows that, the recognition accuracies for the minerals with obvious appearance characteristics such as malachite are higher while those for other minerals like sphalerite are less due to the obfuscation with other minerals. The analysis of the feature map shows that the model extracts the radial feature of malachite perfectly, and the feature vector of mineral image aggregate is in a high degree, which also can prove the reliability of the model.
-
Key words:
- mineral image /
- mineral recognition /
- artificial intelligence /
- deep learning
-
-
表 1 Inception V3模型结构
Table 1. The model structure of Inception V3
处理过程 卷积核大小/步长或者图注 图像大小 备注 conv 3× 3/2 399×399×3 输入矿物图像 conv 3× 3/1 199×199×32 conv padded 3× 3/1 197×197×32 pool 3× 3/2 197×197×64 conv 3× 3/1 98×98×64 conv 3× 3/2 98×98×80 conv 3× 3/1 47×47×192 矿物特征提取 3×Inception 图 1-a 47×47×288 5×Inception 图 1-b 23×23×768 2×Inception 图 1-c 11×11×1280 pool 8×8 11×11×2048 linear logits 1×1×2048 softmax classifier 1×1×1000 输出矿物类别 
下载: 导出CSV
表 2 验证集中矿物识别混淆矩阵
Table 2. Confusion matrix for mineral recognition in validation set
矿物 刚玉 黄铁矿 黄铜矿 金 金红石 孔雀石 蓝铜矿 菱铁矿 绿帘石 闪锌矿 石英 天青石 萤石 自然铜 自然银 方铅矿 刚玉 0.88 0.01 0 0 0.01 0 0 0.02 0 0 0.04 0 0.02 0 0.01 0.01 黄铁矿 0.01 0.85 0.01 0 0.03 0 0 0.01 0 0.03 0.02 0 0 0 0.02 0.02 黄铜矿 0 0.05 0.88 0 0.01 0 0 0 0 0.04 0 0 0 0 0.02 0 金 0 0.01 0.06 0.87 0 0 0 0.04 0 0 0 0 0 0.02 0 0 金红石 0.01 0.03 0 0 0.84 0 0 0 0.01 0 0.04 0 0 0 0.04 0.04 孔雀石 0 0 0 0 0 0.98 0 0 0 0 0 0 0.01 0 0.01 0 蓝铜矿 0 0 0 0 0.01 0 0.98 0 0 0 0 0 0 0 0 0.01 菱铁矿 0.02 0.01 0.06 0 0.02 0 0 0.78 0.01 0.02 0.01 0 0 0.06 0 0.01 绿帘石 0 0.01 0.03 0.01 0 0 0 0.02 0.92 0.01 0 0 0 0 0 0 闪锌矿 0 0.03 0.04 0 0.09 0 0 0.04 0.02 0.74 0 0.02 0 0 0 0.02 石英 0.04 0.01 0 0 0.02 0 0 0 0.01 0 0.85 0 0.06 0 0.01 0 天青石 0.02 0 0 0 0.02 0 0 0.02 0 0.02 0.09 0.75 0.06 0 0.02 0.02 萤石 0.04 0.01 0 0 0 0.02 0 0.01 0 0.02 0.03 0.02 0.86 0 0 0 自然铜 0 0 0.01 0 0.01 0 0 0.05 0 0 0 0 0 0.92 0.01 0 自然银 0 0 0.03 0 0.08 0 0 0 0 0 0 0 0 0.04 0.83 0.02 方铅矿 0.02 0 0.01 0 0.02 0 0.01 0.01 0.01 0.06 0.04 0 0.03 0 0.02 0.76
下载: 导出CSV
-
[1] 方少木.矿物岩石肉眼鉴定[M].北京:煤炭工业出版社, 1980.
[2] 赵鹏大.大数据时代数字找矿与定量评价[J].地质通报, 2015, 34(7):1255-1259. doi: 10.3969/j.issn.1671-2552.2015.07.001 http://dzhtb.cgs.cn/gbc/ch/reader/view_abstract.aspx?file_no=20150701&flag=1
[3] 严光生, 薛群威, 肖克炎, 等.地质调查大数据研究的主要问题分析[J].地质通报, 2015, 34(7):1273-1279. doi: 10.3969/j.issn.1671-2552.2015.07.004 http://dzhtb.cgs.cn/gbc/ch/reader/view_abstract.aspx?file_no=20150704&flag=1
[4] 姚震, 杜子图.关于新时代地质调查工作的战略思考[J].地质通报, 2018, 37(11):2120-2124. http://dzhtb.cgs.cn/gbc/ch/reader/view_abstract.aspx?file_no=20181118&flag=1
[5] 赵明等.矿物学导论[M].北京:地质出版社, 2017.
[6] Mengko T R. Digital Image Processing Technique in Rock Forming Minerals Idenificaiton[C]//IEEE Asia-Pacific Conference on Circuits and Systems, 2000: 4.
[7] Thompson S, Fueten F, Bockus D. Mineral identification using artificial neural networks and the rotating polarizer stage[J]. Computers & Geosciences, 2001, (27):1081-1089. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=74a95d621882825641fc18f6d6dec061
[8] Baykana N A, Yılmaz N. Mineral identification using color spaces and artificial neural networks[J]. Computers & Geosciences, 2010, (36):91-97. http://d.old.wanfangdata.com.cn/NSTLQK/NSTL_QKJJ0224189078/
[9] 叶润青, 牛瑞卿, 张良培, 等.基于图像分类的矿物含量测定及精度评价[J].中国矿业大学学报, 2011, 40(5):810-815+822. http://d.old.wanfangdata.com.cn/Periodical/zgkydxxb201105025
[10] Ghiasi-Freez J, Soleimanpour I, Kadkhodaie-Ilkhchi A, et al. Semi-automated porosity identi fi cation from thin section images using image analysis and intelligent discriminant classifiers[J]. Computers & Geosciences, 2012, (45):36-45. https://www.sciencedirect.com/science/article/pii/S0098300412000945
[11] Aprile A, Castellano G, Eramo G. Combining image analysis and modular neural networks for classification of mineral inclusions and pores in archaeological potsherds[J]. Journal of Archaeological Science, 2014, (50):262-272. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=ef5b9045481f916d49c0d133de0bf346
[12] Aligholi S, Khajav R, Razmara G. Automated mineral identification algorithm using optical properties of crystals[J]. Computers & Geosciences, 2015, (85):175-183. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=c22ec35ccfc479345d11468fea6cb4d3
[13] Baklanova O E, Baklanov A, Shvets O Y. Methods and algorithms of computer vision for automated processing of mineral rocks images[C]//2015 IEEE 10th Jubilee International Symposium on Applied Computational Intelligence and Informatics. Timisoara, 2015: 449-454.
[14] 白林, 姚钰, 李双涛, 等.基于深度学习特征提取的岩石图像矿物成分分析[J].中国矿业, 2018, 27(7):178-182. http://d.old.wanfangdata.com.cn/Periodical/zgky201807038
-
图(6)
表(2)
计量
- 文章访问数: 889
- PDF下载数: 13
- 施引文献: 0