FPN在遥感图像检测中的应用
|
摘要
基于军事领域对遥感图像目标检测技术的需求,研究了深度学习算法中的Faster R-CNN算法,同时针对遥感图像的小目标数目较多,相邻较近等特点,研究了检测算法中的优化算法—FPN算法;在此基础上使用Caffe进行实验仿真,对比了结合不同尺度特征信息的检测模型对遥感图像中飞行器类别的检测结果;试验结果表明,Faster RCNN算法在遥感图像飞行器类别上表现一般,但结合FPN算法思想后检测结果明显提升,最好的检测模型精度提升了8. 7%;基于该检测模型,检测其他种类的遥感目标,只需要对现有的模型进行微调即可;通过优化基础的Faster R-CNN网络结构训练检测模型,能提升检测结果,为军事上的图像检测提供一种新的方向,可以避免传统目标检测过程中需要人工设计特征、检测耗时较长等缺点,也为后续自动目标检测技术的研究提供新的方向。
On the basis of the target detection technology of remote sensing image which needs to be applied in military field,this paper studies Faster R-CNN algorithm in depth learning algorithm. At the same time,in view of the characteristics of large number of small targets and close neighborhood in remote sensing images,the optimization algorithm of detection algorithm-FPN algorithm is studied. On this basis,Caffe is used for experimental simulation,and the detection results of aircraft categories in remote sensing images are compared with the detection models based on different scales of feature information. The experimental results show that the Faster R-CNN algorithm performs well in the classification of aircraft in remote sensing images based on this method. However,the detection results have been significantly improved by combining the idea of FPN algorithm,and the best detection model has been improved by 8. 7%. Based on the detection model in this paper,if we want to detect other kinds of remote sensing targets,we only need to fine-tune the existing models. By optimizing the basic Faster R-CNN network structure training detection model,the detection results can be improved,which provides a new direction for military image detection. It can avoid the shortcomings of traditional target detection process,such as the need for manual design features,long detection time and so on. It also provides a new direction for the follow-up research of automatic target detection technology.
On the basis of the target detection technology of remote sensing image which needs to be applied in military field,this paper studies Faster R-CNN algorithm in depth learning algorithm. At the same time,in view of the characteristics of large number of small targets and close neighborhood in remote sensing images,the optimization algorithm of detection algorithm-FPN algorithm is studied. On this basis,Caffe is used for experimental simulation,and the detection results of aircraft categories in remote sensing images are compared with the detection models based on different scales of feature information. The experimental results show that the Faster R-CNN algorithm performs well in the classification of aircraft in remote sensing images based on this method. However,the detection results have been significantly improved by combining the idea of FPN algorithm,and the best detection model has been improved by 8. 7%. Based on the detection model in this paper,if we want to detect other kinds of remote sensing targets,we only need to fine-tune the existing models. By optimizing the basic Faster R-CNN network structure training detection model,the detection results can be improved,which provides a new direction for military image detection. It can avoid the shortcomings of traditional target detection process,such as the need for manual design features,long detection time and so on. It also provides a new direction for the follow-up research of automatic target detection technology.
引文
[1]Lowe D.G.Object recognition from local scale-invariant features[C].The proceedings of the seventh IEEE international conference on Computer vision.1999,2:1150-1157.
[2]Dalal N.and Triggs B.Histograms of oriented gradients for human detection[C].IEEE Computer Society Conference on Computer Vision and Pattern Recognition.2005,1:886-893.
[3]黄咨.基于卷积神经网络的目标检测模型的研究[D].上海:上海交通大学,2014.
[4]Krizhevsky A,Sutskever I,and Hinton G.Image Net Classification with Deep Convolutional Neural Networks[C].Conference and Workshop on Neural Information Processing Systems,2012.
[5]Image Data Set[DB]:http:∥www.image-net.org/.
[6]Ross Girshick.Fast R-CNN[C].IEEE International Conference on Computer Vision 2015.arXiv:1504.08083.
[7]Shaoqing Ren,Kaiming He,Ross Girshick,Jian Sun.Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks[C].Conference and Workshop on Neural Information Processing Systems,2015.
[8]Kaiming He,Georgia Gkioxari,Piotr Dollar,Ross Girshick.Mask R-CNN[C].IEEE International Conference on Computer Vision,2017.ar Xiv:1703.06870.
[9]Tsung-Yi Lin,Piotr Dollar,Ross Girshick,Kaiming He,Bharath Hariharan,Serge Belongie.Feature Pyramid Networks for Object Detection[C].IEEE Computer Society Conference on Computer Vision and Pattern Recognition,2017.ar Xiv:1612.03144.
[10]Abhinav Shrivastava,Abhinav Gupta,Ross Girshick.Training Region-based Object Detectors with Online Hard Example Mining[C].IEEE Computer Society Conference on Computer Vision and Pattern Recognition,2016.arXiv:1604.03540.
[11]Kaiming He,Xiangyu Zhang,Shaoqing Ren.Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition[C].IEEE Transactions on Pattern Analysis&Machine Intelligence,2015.
[12]常海涛,苟军年,李晓梅.Faster R-CNN在工业CT图像缺陷检测中的应用[J].中国图像图形学报,2018,23(7):1061-1071.
[2]Dalal N.and Triggs B.Histograms of oriented gradients for human detection[C].IEEE Computer Society Conference on Computer Vision and Pattern Recognition.2005,1:886-893.
[3]黄咨.基于卷积神经网络的目标检测模型的研究[D].上海:上海交通大学,2014.
[4]Krizhevsky A,Sutskever I,and Hinton G.Image Net Classification with Deep Convolutional Neural Networks[C].Conference and Workshop on Neural Information Processing Systems,2012.
[5]Image Data Set[DB]:http:∥www.image-net.org/.
[6]Ross Girshick.Fast R-CNN[C].IEEE International Conference on Computer Vision 2015.arXiv:1504.08083.
[7]Shaoqing Ren,Kaiming He,Ross Girshick,Jian Sun.Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks[C].Conference and Workshop on Neural Information Processing Systems,2015.
[8]Kaiming He,Georgia Gkioxari,Piotr Dollar,Ross Girshick.Mask R-CNN[C].IEEE International Conference on Computer Vision,2017.ar Xiv:1703.06870.
[9]Tsung-Yi Lin,Piotr Dollar,Ross Girshick,Kaiming He,Bharath Hariharan,Serge Belongie.Feature Pyramid Networks for Object Detection[C].IEEE Computer Society Conference on Computer Vision and Pattern Recognition,2017.ar Xiv:1612.03144.
[10]Abhinav Shrivastava,Abhinav Gupta,Ross Girshick.Training Region-based Object Detectors with Online Hard Example Mining[C].IEEE Computer Society Conference on Computer Vision and Pattern Recognition,2016.arXiv:1604.03540.
[11]Kaiming He,Xiangyu Zhang,Shaoqing Ren.Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition[C].IEEE Transactions on Pattern Analysis&Machine Intelligence,2015.
[12]常海涛,苟军年,李晓梅.Faster R-CNN在工业CT图像缺陷检测中的应用[J].中国图像图形学报,2018,23(7):1061-1071.