轻量化自适应特征选择目标检测网络

doi:10.12068/j.issn.1005-3026.2021.09.004

摘要/Abstract

摘要： 在小目标物体检测、多类别物体检测尤其是轻量化检测模型等关键技术研究方面仍面临较大的挑战，基于此，本文提出一种轻量化自适应特征选择目标检测网络.该网络以特征金字塔为基础，提取多尺度图像特征并从空间维度上对特征图进行滤波，从通道维度上自适应地选择特征图中更重要的通道，降低多通道下噪声和干扰对目标特征的稀释作用，减少特征图在传递过程中的信息丢失.除此之外，构建深度可分离卷积的分类网络，降低后续处理的计算量，加快检测速度，实现网络的轻量化处理.在PASCAL VOC 2007数据集上的检测平均精度为77.7%，检测速度为14.3帧/s.在MS COCO数据集上的测试结果表明，该网络在精度损失小于5%的情况下，检测速度远超FPN，比Mask R-CNN可以更好地兼顾检测速度和检测精度.

关键词: 目标检测;特征金字塔;自适应特征选择;轻量化网络

Abstract: There are some limitations of key technologies in small object detection， multi-category object detection， and especially lightweight models. To solve these problems， this paper proposes a lightweight adaptive feature selection network for object detection. The network is constructed to extract multi-scale features based on the feature pyramid. To alleviate the noise interference and preserve the detail information， a feature selection module composed of spatial adaptation and channel adaptation is designed. Specifically， the feature maps are filtered from spatial dimension， encoded and decoded in channel dimension for selecting the meaningful features adaptively. Besides， the classification network is constructed through depth-wise separable convolution to reduce the computational cost， improve detection efficiency and realize the lightweight version of the network. The detection accuracy on the PASCAL VOC 2007 dataset is 77.7% in mean average precision(mAP)， and the detection speed is 14.3 in frames per second(FPS). The results on the MS COCO dataset show that the proposed network outperforms FPN and Mask R-CNN at the cost of 5% accuracy loss and achieves a better balance between accuracy and efficiency.

Key words: object detection; feature pyramid; adaptive feature selection; lightweight network

中图分类号:

TP391

杨爱萍，宋尚阳，程思萌. 轻量化自适应特征选择目标检测网络[J]. 东北大学学报（自然科学版）, 2021, 42(9): 1238-1245.

YANG Ai-ping， SONG Shang-yang， CHENG Si-meng. Lightweight Adaptive Feature Selection Network for Object Detection[J]. Journal of Northeastern University(Natural Science), 2021, 42(9): 1238-1245.

参考文献

[1]Zitnick C L，Dollár P.Edge boxes:locating object proposals from edges［C］//Proceedings of the European Conference on Computer Vision.Berlin:Springer-Verlag，2014:391-405.
[2]Uijlings J R R，Van de Sande K E A，Gevers T，et al.Selective search for object recognition［J］.International Journal of Computer Vision，2013，104(2):154-171.
[3]Lowe D G.Distinctive image features from scale-invariant keypoints［J］.International Journal of Computer Vision，2004，60(2):91-110.
[4]Joachims T.A probabilistic analysis of the rocchio algorithm with TFIDF for text categorization［C］// Proceedings of the International Conference on Machine Learning.San Francisco，1997:143-151.
[5]Dalal N，Triggs B.Histograms of oriented gradients for human detection［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.San Diego，2005:886-893.
[6]Freund Y，Schapire R E.A decision-theoretic generalization of on-line learning and an application to boosting［J］.Journal of Computer and System Sciences，1997，55(1):119-139.
[7]Cortes C，Vapnik V.Support-vector networks［J］.Machine Learning，1995，20(3):273-297.
[8]Redmon J，Divvala 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.Las Vegas，2016:779-788.
[9]Liu W，Anguelov D，Erhan D，et al.SSD:single shot multibox detector［C］//Proceedings of the European Conference on Computer Vision.Berlin:Springer-Verlag，2016:21-37.
[10]Zhang S，Wen L，Bian X，et al.Single-shot refinement neural network for object detection［C］// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Salt Lake City，2018:4203-4212.
[11]Zhu C，He Y，Savvides M.Feature selective anchor-free module for single-shot object detection［C］// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Long Beach，2019:840-849.
[12]Girshick R，Donahue J，Darrell T，et al.Rich feature hierarchies for accurate object detection and semantic segmentation［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Columbus，2014:580-587.
[13]Girshick R.Fast R-CNN［C］//Proceedings of the IEEE International Conference on Computer Vision.Santiago，2015:1440-1448.
[14]Ren S，He K，Girshick R，et al.Faster R-CNN:towards real-time object detection with region proposal networks［C］//Proceedings of the Neural Information Processing Systems.Montréal，2015:91-99.
[15]Dai J，Li Y，He K，et al.R-FCN:object detection via region-based fully convolutional networks［C］//Proceedings of the Neural Information Processing Systems.Barcelona，2016:379-387.
[16]He K M，Gkioxari G，Dollár P，et al.Mask R-CNN［C］// Proceedings of the IEEE International Conference on Computer Vision.Venice，2017:2980-2988.
[17]Lin T Y，Dollár P，Girshick R，et al.Feature pyramid networks for object detection［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Honolulu，2017:936-944.
[18]Pang J，Chen K，Shi J，et al.Libra R-CNN:towards balanced learning for object detection［C］// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Long Beach，2019:821-830.
[19]Sandler M，Howard A，Zhu M，et al.Mobilenetv2:inverted residuals and linear bottlenecks［C］// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Salt Lake City，2018:4510-4520.
[20]Hu J，Shen L，Sun G.Squeeze-and-excitation networks［C］// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Salt Lake City，2018:7132-7141.
[21]Wang X，Cai Z，Gao D，et al.Towards universal object detection by domain attention［C］// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Long Beach，2019:7281-7290.
[22]Everingham M，Van Gool L，Williams C K I，et al.The pascal visual object classes(VOC)challenge［J］.International Journal of Computer Vision，2010，88(2):303-338.
[23]Lin T Y，Maire M，Belongie S，et al.Microsoft COCO:common objects in context［C］//Proceedings of the European Conference on Computer Vision.Berlin:Springer-Verlag，2014:740-755.
[24]Redmon J，Farhadi A.YOLO9000:better，faster，stronger［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Honolulu，2017:6517-6525.
[25]Kong T，Sun F，Yao A，et al.RON:reverse connection with objectness prior networks for object detection［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Honolulu，2017:5244-5252.
[26]Fu C Y，Liu W，Ranga A，et al.Dssd:deconvolutional single shot detector［J］.arXiv Preprint arXiv，2017:1701.06659.
[27]Redmon J，Farhadi A.Yolov3:an incremental improvement［J］.arXiv Preprint arXiv，2018:1804.02767.
[28]Bell S，Lawrence Z C，Bala K，et al.Inside-outside net:detecting objects in context with skip pooling and recurrent neural networks［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas，2016:2874-2883.
[29]Shrivastava A，Gupta A，Girshick R.Training region-based object detectors with online hard example mining［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas，2016:761-769.(上接第1225页)
[12]Torelli F，Vaccaro A.A generalized computing paradigm based on artificial dynamic models for mathematical programming［J］.Soft Computing，2014，18(8):1561-1573.
[13]Torelli F，Vaccaro A.A second order dynamic power flow model［J］.Electric Power Systems Research，2015，126(1):12-20.
[14]Liang X，Mosalam K M.Lyapunov-based nonlinear solution algorithm for structural analysis［J］.Journal of Engineering Mechanics，2018，144(9):12-21.
[15]Chellaboina V，Leonessa A，Haddad W M.Generalized Lyapunov and invariant set theorems for nonlinear dynamical systems［J］.Systems & Control Letters，1999，38(45):289-295.
[16]Xie N，Bompard E，Napoli R，et al.Widely convergent method for finding solutions of simultaneous nonlinear equations［J］.Electric Power Systems Research，2012，83(1):9-18.
[17]Xie N，Torelli F，Bompard E，et al.Dynamic computing paradigm for comprehensive power flow analysis［J］.IET Generation Transmission & Distribution，2013，7(8):832-842.
[18]Fonseca C M，Fleming P J.Multiobjective optimization and multiple constraint handling with evolutionary algorithms.Part I: unified formulation［J］.IEEE Transactions on Systems Man and Cybernetics.Part A-Systems and Humans，1998，28(1):26-37.