基于Mask R-CNN的铁谱磨粒智能分割与识别  

Ferrographic Wear Debris Intelligent Segmentation and Recognition Based on Mask R-CNN

作　　者：安超 魏海军[1] 刘竑[1] 梁麒立 汪璐璐 AN Chao;WEI Haijun;LIU Hong;LIANG Qili;WANG Lulu(Merchant Marine College,Shanghai Maritime University,Shanghai 201306,China;College of Information Engineering,Shanghai Maritime University,Shanghai 201306,China)

机构地区：[1]上海海事大学商船学院,上海201306 [2]上海海事大学信息工程学院,上海201306

出　　处：《润滑与密封》2020年第3期107-112,共6页Lubrication Engineering

基　　金：上海市科学技术委员会资助项目(17ZR1412700).

摘　　要：针对铁谱图像因背景复杂、尺寸分布广、颗粒重叠等导致难以精确分割与识别的问题,以相似度高的疲劳剥块、严重滑动磨粒、层状磨粒共3种异常磨粒作为研究对象,提出基于深度神经网络模型Mask R-CNN的对多目标铁谱磨粒进行智能分割与识别的方法,并对特征提取层分别选用深度不同的残差网络ResNet50和ResNet101进行对比试验。实验结果表明,基于迁移学习方法的Mask R-CNN+ResNet101模型能够在复杂背景下对多目标、多类型、多尺寸的相似磨粒进行有效分割与识别,测试集的平均精度高达76.2%,模型具有较好的泛化能力。It is difficult to segment and recognize ferrography image precisely due to the complex background,wide size distribution and overlapping debris.An intelligent multi-target wear debris segmentation and recognition method based on the deep neural network model Mask R-CNN was propose to study three kinds of abnormal abrasive,including fatigue spall,severe sliding debris,laminar debris.For feature extraction layer,residual network ResNet50 and ResNet101 with different depths were selected for comparative test.The experimental results show that Mask R-CNN+ResNet101 can effectively segment and identify ferrographic wear debris of multiple targets,types and sizes under complex background.The average precision of the test set is as high as 76.2%,and the model has good generalization ability.

关 键 词：深度神经网络 铁谱磨粒 迁移学习 MASK R-CNN 分割与识别 

分 类 号：TH117.2[机械工程—机械设计及理论]