摘要
根据离子吸附型稀土矿开采过程中的沉淀池状态及其空间分布关系,构建了基于Mask R-CNN的高分遥感影像稀土开采识别方法,实现稀土开采状态的识别与检测,为离子型稀土开采监管提供技术支持.该方法采用ROI Align结合双线性插值精确计算特征图浮点数处的像素值,避免了ROI Pooling带来的检测框偏离问题;同时在算法中引入特征金字塔网络(FPN),提升了算法对多尺度下小目标的精准检测能力;针对离子型稀土开采过程中沉淀池存在浸矿液体特征,加入归一化水体指数(NDWI)作为样本训练模型,再将模型用于离子型稀土开采识别;对比了Mask R-CNN,Faster R-CNN,SSD和YOLOv3检测算法在不同的基础网络和数据集下对目标的识别效果.结果表明:4种算法在使用NDWI数据集作为训练样本时比使用原始RGB数据集时的目标检测精度均提升了1.5%左右,且Mask R-CNN的识别效果优于其他3种算法.
According to the sedimentation tank state in the process of rare earth mining and the spatial distribution, a rare earth mining identification algorithm based on Mask R-CNN was proposed to realize the identification and detection of rare earth mining state. ROI Align combined with bilinear interpolation was used to accurately calculate the pixel value at the floating point of the feature graph to avoid the detection frame deviation caused by RoI Pooling. At the same time, a feature pyramid networks were introduced into the algorithm to improve the ability to accurately detect small targets at multiple scales. In addition, for the characteristics of leaching liquid in the sedimentation tank during the ion rare earth mining process, the normalized difference water index(NDWI) was added as the sample to train the model, and then the model was applied to the recognition of rare earth mining. Finally, the recognition effects among Mask R-CNN, Faster R-CNN, SSD and YOLOv3 detection algorithms were compared under different basic networks and data set. The results show that all of the four algorithms improve the target detection accuracy by about 1.5% when using the NDWI data set as training samples compared to using the original RGB data set, and the accuracy of Mask R-CNN recognition is better than the other three algorithms.
作者
李恒凯
肖松松
王秀丽
柯江晨
LI Hengkai;XIAO Songsong;WANG Xiuli;KE Jiangchen(College of Architecture and Surveying Engineering,Jiangxi University of Science and Technology,Ganzhou,Jiangxi 341000,China;College of Economic Management,Jiangxi University of Science and Technology,Ganzhou,Jiangxi 341000,China)
出处
《中国矿业大学学报》
EI
CAS
CSCD
北大核心
2020年第6期1215-1222,共8页
Journal of China University of Mining & Technology
基金
教育部人文社会科学研究规划项目(18YJAZH040)
江西省自然科学基金项目(20181BAB206018)
江西省教育厅科学技术研究重点项目(GJJ180423)。
