黄晓红, 李铁锋, 刘祥鑫,等.基于改进Faster R-CNN算法的岩石裂隙发展方向跟踪预测研究[J].河南理工大学学报（自然科学版）,2022,41(4):134-141.

HUANG X H, LI T F, LIU X X, et al.Research on tracking and prediction of rock fissure development based on improved Faster R-CNN algorithm[J].Journal of Henan Polytechnic University(Natural Science) ,2022,41(4):134-141.

基于改进Faster R-CNN算法的岩石裂隙发展方向跟踪预测研究

黄晓红1, 李铁锋1, 刘祥鑫2, 李伟1

1.华北理工大学 人工智能学院，河北 唐山063210;2.华北理工大学 矿业工程学院，河北 唐山 063210

摘要:为了避免部分岩土工程灾害发生，对岩石表面裂隙发展方向进行跟踪预测，提高岩石表面裂隙的检出率，提出一种基于改进Faster R-CNN算法的岩石裂隙发展方向跟踪预测方法。该法针对红外热像图中岩石裂隙形态多变、尺寸差异大并对实时性有一定要求的特点，以深度残差网络ResNet50为特征提取网络，利用ROIAlign优化建议框与岩石裂隙特征的空间对应关系，建立特征金字塔融合多尺度特征，对Faster R-CNN算法进行改进，并结合平均红外辐射度-时间曲线对岩石裂隙发展方向跟踪预测，使用花岗岩单轴压缩试验中采集的红外光谱特征 进行试验测试。结果表明，该方法能够较好检测红外热像图中的岩石裂隙，在测试集上的 mAP达到88.81%，泛化能力较强，同时结合检测框内的平均红外辐射温度-时间曲线可以较好地跟踪预测裂隙发展方向。

关键词:岩石裂隙;红外热像图;平均红外辐射温度;目标检测;改进FasterR-CNN算法

doi:10.16186/j.cnki.1673-9787.2020100031

基金项目:国家自然科学基金资助项目（41977219）；河北省高等学校科学技术重点研究项目（ZD2020152）

收稿日期:2020/10/15

修回日期:2021/03/15

出版日期:2022/07/15

Research on tracking and prediction of rock fissure development based on improved Faster R-CNN algorithm

HUANG Xiaohong1, LI Tiefeng1, LIU Xiangxin2, LI Wei1

1.College of Artificial Intelligence，North China University of Science and Technology，Tangshan 063210，Hebei，China;2.College of Mining Engineering，North China University of Science and Technology，Tangshan  063210，Hebei，China

Abstract: In order to track and predict the development of rock surface cracks so as to avoid the occurrence of some geotechnical disasters and improve the detection rate of rock surface fissures, a method for tracking and predicting rock fissure development based on improved Faster R-CNN algorithm was proposed. Aiming at the characteristics of rock fissures in the infrared thermal image that the shape of the rock fissures were variable, the size difference was large,and there were certain requirements for real-time performance , the deep residual network ResNet50was used as the feature extraction network , and ROIAlign was used to optimize the space between the proposed frame and the rock fissure feature. A feature pyramid was established to fuse multi-scale features to improve the Faster R-CNNalgorithm, and the average infrared radiation temperature-time curve was used to track and predict the development of rock fractures. The infrared features collected in the uniaxial compression experiment of granite were used for experimental testing. The results showed that this method had a good detection effect on the rock fissures in the infrared thermal image. The mAP on the test set reached 88. 81%，and the generalization ability was strong. Compared with other common target detection methods，the detection effect was better. At the same time，combined with the average infrared radiation temperature-time curve in the detection frame，it had a certain effect on the tracking and prediction of the crack development.

Key words:rock fissure;thermal infrared image;average infrared radiation temperature;target detection;improved Faster R-CNN algorithm

基于改进Faster_R-...石裂隙发展方向跟踪预测研究_黄晓红.pdf