基于SSD与图像变换的镜下矿物光片智能识别

doi:10.12068/j.issn.1005-3026.2025.20230347

摘要/Abstract

摘要：

在矿物识别中，当识别伴生矿物时，有时会产生漏判、误判.为了解决上述问题，开展了显微镜下矿物的智能化识别方法研究.首先，改进了SSD网络并结合图像变换构建了一种智能识别方法；其次，将该方法应用于中国辽宁省某铁矿光片的显微镜下矿物图像，通过试验证明了方法的准确性；最后，确定了学习率、批量尺寸对损失函数的影响，使用梯度下降法进一步提高了识别精度.在试验中，识别精度超过90%，最高可达100%，损失函数值最小值约为0.008.结果表明，提出的方法具有较强的矿物识别能力.

关键词: 矿物识别, 深度学习, SSD, 图像变换, 矿物含量估算

Abstract:

In mineral identification， there may sometimes be cases of missed identification or incorrect identification when identifying associated minerals. To address these issues， the intelligent mineral identification methods under the microscope were developed. Firstly， an intelligent identification method was constructed by improving the SSD network and incorporating image transformation. Secondly， the method was applied to microscope images of minerals in polished sections from the iron ore in Liaoning Province， China， and its accuracy was demonstrated through tests. Finally， the effects of learning rate and batch size on the loss function were determined， and the accuracy was further improved by using the gradient descent method. In the tests， the identification accuracy exceeds 90%， reaching up to 100%， with the minimum value of the loss function was approximately 0.008. The results indicate that the proposed method has strong mineral identification capabilities.

Key words: mineral identification, deep learning, single shot multibox detector （SSD）, image transformation（IT）, mineral content estimation

中图分类号:

P 575

侯振隆, 申晋容, 魏继康, 赵文天. 基于SSD与图像变换的镜下矿物光片智能识别[J]. 东北大学学报（自然科学版）, 2025, 46(6): 131-137.

Zhen-long HOU, Jin-rong SHEN, Ji-kang WEI, Wen-tian ZHAO. Intelligent Identification of Minerals Polished Section Under Microscope Based on SSD and Image Transformation[J]. Journal of Northeastern University(Natural Science), 2025, 46(6): 131-137.

图/表 9

图1 改进的SSD网络结构

Fig.1 Network structure of improved SSD

图2 SSD-IT流程示意图

Fig.2 Flow chart of SSD-IT

图3 试验数据增强示意图（a）—待增强图像数据；（b）—镜像处理后的图像数据；（c）—模糊处理后的图像数据；（d）—缩放处理后的图像数据；（e）—亮度变化后的图像数据；（f）—旋转处理后的图像数据；（g）—随机裁剪后的图像数据.

Fig.3 Schematic diagram of test data enhancement

图4 SSD-IT图像数据标注方式与识别结果示意图（a）—标注的矿物图像；（b）—识别结果.

Fig.4 Data labeling and identification results of SSD-IT image

表1 不同识别模型的评价结果 (%)

Table 1 Evaluation results of different identification models

模型	识别精度	召回率	F0.5-score	F1-score	F2-score
SSD-IT	91.19	80.91	88.93	85.74	82.78
SSD	24.46	40.00	26.33	30.36	35.40
Faster-RCNN-IT	40.99	52.62	42.88	46.08	49.79
Faster-RCNN	9.63	17.78	10.60	12.49	15.21
YOLOv5s-IT	40.30	1.45	6.33	2.80	1.80
YOLOv5s	11.80	4.35	8.79	6.36	4.98

图5 不同光片的镜下图像识别精度

Fig.5 Identification precision for the microscopic images from polished sections

图6 不同学习率情况下的损失值（a）—不同学习率下损失值曲线图；（b）—学习率恒定为0.010以及学习率随迭代次数变化的损失值曲线图.

Fig.6 Loss values for different learning rates

图7 不同批量大小情况下的损失值

Fig.7 Loss values for different batch sizes

图8 图像变换前后的对比（a）—变换前；（b）—变换后.

Fig.8 Comparison of the images before and after the transformation

参考文献 23

[1]	Młynarczuk M， Górszczyk A， Bartłomiej Ś. The application of pattern recognition in the automatic classification of microscopic rock images ［J］. Computers & Geosciences， 2013， 60： 126-133.
[2]	Xu Z H， Ma W， Lin P， et al. Deep learning of rock microscopic images for intelligent lithology identification： neural network comparison and selection ［J］. Journal of Rock Mechanics and Geotechnical Engineering， 2022， 14（4）： 1140-1152.
[3]	Polat Ö， Polat A， Ekici T. Automatic classification of volcanic rocks from thin section images using transfer learning networks ［J］. Neural Computing and Applications， 2021， 33（18）： 11531-11540.
[4]	Zhou Y M， Wong L N Y， Tse K K C. Novel rock image classification： the proposal and implementation of HKUDES_Net ［J］. Rock Mechanics & Rock Engineering， 2023， 56（5）： 3825-3841.
[5]	Singh V， Rao S M. Application of image processing and radial basis neural network techniques for ore sorting and ore classification ［J］. Minerals Engineering， 2005， 18（15）： 1412-1420.
[6]	Tessier J， Duchesne C， Bartolacci G. A machine vision approach to on-line estimation of run-of-mine ore composition on conveyor belts ［J］. Minerals Engineering， 2007， 20（12）： 1129-1144.
[7]	Ding K， Xue L F， Ran X J， et al. Siamese network based prospecting prediction method： a case study from the Au deposit in the Chongli mineral concentrate area in Zhangjiakou， Hebei Province， China ［J］. Ore Geology Reviews， 2022， 148： 105024.
[8]	Wang X P， Zuo R G， Wang Z Y. Lithological mapping using a convolutional neural network based on stream sediment geochemical survey data ［J］. Natural Resources Research， 2022， 31（5）： 2397-2412.
[9]	谢洪途，陈佳兴，张琳，等.基于脉冲神经网络的轻量化SAR图像舰船识别算法［J］.东北大学学报（自然科学版），2024，45（4）：474-482.
	Xie Hong-tu， Chen Jia-xing， Zhang Lin， et al. Lightweight ship recognition algorithm based on SNN in SAR images ［J］. Journal of Northeastern University（Natural Science），2024，45（4）：474-482.
[10]	武锦龙，吴虹麒，李浩，等.基于改进DeepLabV^{3 +} 的荞麦苗期无人机遥感图像分割识别方法研究［J］.农业机械学报，2024，55（5）：186-195.
	Wu Jin-long， Wu Hong-qi， Li Hao， et al. Segmentation of buckwheat by UAV based on improved lightweight DeepLabV^{3 +} at seedling stage ［J］. Transactions of the Chinese Society for Agricultural Machinery， 2024，55（5）：186-195.
[11]	Chen J Y， Yang T J， Zhang D M， et al. Deep learning based classification of rock structure of tunnel face ［J］. Geoscience Frontiers， 2021， 12（1）： 395-404.
[12]	Deo A J， Sahoo A， Behera S K， et al. Vision-based size classification of iron ore pellets using ensembled convolutional neural network ［J］. Neural Computing and Applications， 2022， 34（21）： 18629-18641.
[13]	Zhou W Y， Wang H， Wan Z B. Ore image classification based on improved CNN ［J］. Computers and Electrical Engineering， 2022， 99： 107819.
[14]	王伟，李擎，张德政，等.基于深度学习的矿石图像处理研究综述［J］.工程科学学报，2023，45（4）：621-631.
	Wang Wei， Li Qing， Zhang De-zheng， et al. A survey of ore image processing based on deep learning ［J］. Chinese Journal of Engineering， 2023， 45（4）： 621-631.
[15]	Ren S Q， He K M， 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.
[16]	Redmon J， Divvala S， Girshick R， et al. You only look once： unified， real-time object detection ［C］ // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. Las Vegas， 2016： 779-788.
[17]	Liu W， Anguelov D， Erhan D， et al. SSD： single shot multibox detector ［C］ // ECCV 2016： 14th European Conference. Amsterdam， 2016： 21-37.
[18]	Zhang S F， Wen L Y， Bian X， et al. Single-shot refinement neural network for object detection ［C］ //2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. Salt Lake City， 2018： 4203-4212.
[19]	Lin T Y， Dollár P， Girshick R， et al. Feature pyramid networks for object detection ［C］ // 2017 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. Honolulu，2017： 2117-2125.
[20]	He K M， Zhang X Y， Ren S Q， et al. Deep residual learning for image recognition ［C］ // 2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. Las Vegas， 2016： 770-778.
[21]	Pratt W K. Digital image processing： PIKS scientific inside ［M］. Hoboken： Wiley-Interscience， 2007.
[22]	Chinchor N， Sundheim B. MUC-5 evaluation metrics ［C］ // Proceedings of the 5th Conference on Message Understanding—MUC 5’93. Baltimore， 1993： 69-78.
[23]	Frankes W B， Baeza-Yates R. Information retrieval： data structure & algorithms ［M］. Englewood cliffs： PrenticeHall， 1992.