一种改进的HED网络及其在边缘检测中的应用
|
摘要
HED网络(holistically nested edge detection network)被证明是目前用于边缘检测的一种性能较好的深度学习网络,但在实际应用中发现,将该网络用于前下视红外成像制导自动目标识别时,会出现检测出的边缘不完整、不光滑等问题。针对上述问题,对HED网络进行了改进,在此基础上提出了一种基于改进HED网络的边缘提取方法。首先,在原网络结构的基础上减少了两个池化层,提高了侧边输出层的输出精度,然后,将改进HED网络输出的边缘概率图进行二值化,得到显著性边缘区域;最后,采用基于匹配滤波的边缘提取方法提取图像的边缘,并将其与改进HED网络提取的显著目标边缘相融合,得到最终结果。实验结果表明,该方法能够大幅减少非目标区域的边缘,并且能够有效提取较为完整和准确的目标边缘,为后续红外图像中的目标检测、跟踪与识别奠定了良好的基础。
The holistically nested edge detection(HED) network has proven to be a better deep-learning network for edge detection so far. However, it is found that the detected edges are incomplete and not smooth when the network is used for automatic target recognition in forward-looking infrared imaging guidance. To solve these problems, the HED network needs to be improved. Hence, an edge extraction method based on the improved HED network is proposed. First, two pooling layers are reduced on the basis of the original HED network structure, and the output accuracy of the side output layer is improved. Then, the edge probability map of the output of the improved HED network is binarized to obtain a significant edge region. Finally, the edge extracted by a matching filter is merged with the edge extracted by the improved HED network. The experimental results show that this method can significantly reduce the edges of non-target areas, and can effectively extract relatively complete and accurate target edges, laying a good foundation for target detection, tracking, and recognition in subsequent infrared images.
The holistically nested edge detection(HED) network has proven to be a better deep-learning network for edge detection so far. However, it is found that the detected edges are incomplete and not smooth when the network is used for automatic target recognition in forward-looking infrared imaging guidance. To solve these problems, the HED network needs to be improved. Hence, an edge extraction method based on the improved HED network is proposed. First, two pooling layers are reduced on the basis of the original HED network structure, and the output accuracy of the side output layer is improved. Then, the edge probability map of the output of the improved HED network is binarized to obtain a significant edge region. Finally, the edge extracted by a matching filter is merged with the edge extracted by the improved HED network. The experimental results show that this method can significantly reduce the edges of non-target areas, and can effectively extract relatively complete and accurate target edges, laying a good foundation for target detection, tracking, and recognition in subsequent infrared images.
引文
[1]肖传民,史泽林,夏仁波,等.一种基于视觉显著性的边缘检测算法[J].信息与控制,2014,43(1):9-13.XIAO Chuanmin,SHI Zelin,XIA Renbo,et al.Edge detection algorithm based on visual saliency[J].Information and Control,2014,43(1):9-13.
[2]王晨,汤心溢,高思莉.基于深度卷积神经网络的红外场景理解算法[J].红外技术,2017,39(8):728-733.WANG Chen,TANG Xinyi,GAO Sili.Infrared Scene Understanding Algorithm Based on Deep Convolutional Neural Network[J].Infrared Technology,2017,39(8):728-733.
[3]SHEN W,WANG X,WANG Y,et al.Deep Contour:A deep convolutional feature learned by positive-sharing loss for contour detection[C]//Computer Vision and Pattern Recognition.IEEE,2015:3982-3991.
[4]Bertasius G,Shi J,Torresani L.High-for-Low and Low-for-High:Efficient Boundary Detection from Deep Object Features and Its Applications to High-Level Vision[C]//IEEE International Conference on Computer Vision.IEEE,2015:504-512.
[5]XIE S,TU Z.Holistically-Nested Edge Detection[J].International Journal of Computer Vision,2015:1-16.
[6]Simonyan K,Zisserman A.Very Deep Convolutional Networks for Large-Scale Image Recognition[J].Computer Science,2014.
[7]LUO H B,JIAO A B,XU L Y,et al.Edge detection using matched filter[C]//Control and Decision Conference.IEEE,2015:1132-1136.
[8]LIU Y,CHENG M M,HU X,et al.Richer Convolutional Features for Edge Detection[C]//IEEE Conference on Computer Vision and Pattern Recognition.IEEE Computer Society,2017:5872-5881.
[2]王晨,汤心溢,高思莉.基于深度卷积神经网络的红外场景理解算法[J].红外技术,2017,39(8):728-733.WANG Chen,TANG Xinyi,GAO Sili.Infrared Scene Understanding Algorithm Based on Deep Convolutional Neural Network[J].Infrared Technology,2017,39(8):728-733.
[3]SHEN W,WANG X,WANG Y,et al.Deep Contour:A deep convolutional feature learned by positive-sharing loss for contour detection[C]//Computer Vision and Pattern Recognition.IEEE,2015:3982-3991.
[4]Bertasius G,Shi J,Torresani L.High-for-Low and Low-for-High:Efficient Boundary Detection from Deep Object Features and Its Applications to High-Level Vision[C]//IEEE International Conference on Computer Vision.IEEE,2015:504-512.
[5]XIE S,TU Z.Holistically-Nested Edge Detection[J].International Journal of Computer Vision,2015:1-16.
[6]Simonyan K,Zisserman A.Very Deep Convolutional Networks for Large-Scale Image Recognition[J].Computer Science,2014.
[7]LUO H B,JIAO A B,XU L Y,et al.Edge detection using matched filter[C]//Control and Decision Conference.IEEE,2015:1132-1136.
[8]LIU Y,CHENG M M,HU X,et al.Richer Convolutional Features for Edge Detection[C]//IEEE Conference on Computer Vision and Pattern Recognition.IEEE Computer Society,2017:5872-5881.