摘要
针对传统零件识别方法图像特征提取鲁棒性不足,零件识别准确率较低、不能对图像进行实例分割的问题,文章提出了一种基于Mask R-CNN的零件识别方法。该方法利用卷积神经网络对零件图像进行特征提取,选取数据集中标注好的图像微调Mask R-CNN网络,以保证零件识别的准确性,并生成Mask分割掩码,对零件进行实例分割。同时,对数据集进行数据增强和划分K折交叉验证来提高模型的鲁棒性。最后通过搭建实验平台对零件进行识别,证明了该方法的有效性。
Aiming at the problem that the image feature extraction of traditional part recognition method is not robust enough, the accuracy of part recognition is low, and the image cannot be segmented by example. This paper proposes a part recognition method based on Mask R-CNN. The method uses the convolutional neural network to extract the features of the part image, selects the annotated part image from dataset to fine-tuned Mask R-CNN to guarantee the accuracy of the part recognition, and generates the segmentation mask to segment the parts. Data expansion and K-Fold Cross Validation also were used to improve the robustness of the model. Finally, the result of part recognition by building the experimental platform proves the effectiveness of the method.
引文
[1] 张青,邹湘军,林桂潮,等.基于高斯混合模型的零件精确识别与定位算法[J/OL].计算机集成制造系统,http://kns.cnki.net/kcms/ detail/11.5946.TP.20180330.1145.004.html,2018-07-29.
[2] 汪嘉杰,王磊,范秀敏,等.基于视觉的航天电连接的智能识别与装配引导[J].计算机集成制造系统,2017,23(11):2423-2430.
[3] 夏庆观,盛党红,路红,等.零件图像特征提取和识别的研究[J].中国机械工程,2005,16(22):2031-2033.
[4] 何晓阳,徐惠钢,谢启.基于LabVIEW与BP神经网络的零件识别系统[J].仪表技术与传感器,2017(1):119-122.
[5] 司小婷,吴文江,孙一兰.基于视觉的零件识别和定位[J].组合机床与自动化加工技术,2016(10):70-73.
[6] Hinton G E,Osindero S,The Y A.Fast learning algorithm for deep belief nets[J].Neural Computation,2006,18:1527-1554.
[7] Redmon J,Divvals 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,2016:779—788.
[8] Liu W,Anguelov D,Erhan D,et al.SSD:Single Shot MultiBox Detector[C]// European Conference on Computer Vision,2016:21-37.
[9] Ren S,He K,Girshick R,et al.Faster R-CNN:towards real-time object detection with region proposal networks[C]// International Conference on Neural Information Processing Systems.MIT Press,2015:91-99.
[10] He K,Gkioxari G,Dollar P,et al.Mask R-CNN[J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2017(99):1.