Object Detection Method Based on Feature Channel Modeling

doi:10.19678/j.issn.1000-3428.0061872

Abstract

Abstract: Addressing the problem that using multiscale fusion feature maps to predict objects without channel modeling results in poor robustness.A detection method based on the multidimensional modeling of global image information is proposed.In this method, multistage feature reuse and feature fusion are used to reduce the loss of correlation within features.Breadth Channel Modeling Branch(BCMB) and Depth Channel Modeling Branch(DCMB) are designed to compensate for the lack of spatial information caused by the change in the receptive field and enrich the contextual information among the objects.The former established a two-dimensional channel matrix in the width and height directions, which can model the multilevel receptive field, thus enriching the spatial perceptual ability of the model and facilitating localization.The latter establishes a one-dimensional channel vector in the depth direction, which can extract the global features of an image, thus enriching the contextual description of the model and helping classification.According to the experimental results, by weighted fusion of the channel map obtained by the two branches with the input feature, the output feature can be sensitive to the position and category information of the object, which makes the mAP value reach 85.8% in the PASCAL VOC 2007 test data set.Compared with the Baseline method without channel modeling, the proposed method improved the mAP value by a maximum of 3.2 percentage points.

Key words: object detection, contextual information, channel modeling, Convolutional Neural Networks(CNN), feature enhancement

摘要： 针对直接利用多尺度融合特征图进行目标检测时鲁棒性较差的问题，提出一种对图像全局信息进行多维建模的检测方法。采用多阶段的特征复用和特征融合减少特征间相关性损失，设计广度通道建模分支（BCMB）与深度通道建模分支（DCMB）弥补因感受野变化造成的图像空间信息不足，并丰富图像中各个目标间的上下文信息。通过BCMB建立宽高方向的二维通道矩阵，对多层级的感受野进行建模，进而丰富模型对图像的空间感知，完成目标定位。使用DCMB建立深度方向的一维通道向量，提炼图像的全局特征，丰富模型对图像的上下文描述，完成目标分类。将2个分支生成的通道图与输入特征进行加权融合，增强图像通道表达力，使输出的特征对目标的位置和类别信息更敏感。在PASCAL VOC 2007测试数据集上的实验结果表明，该方法的mAP值为85.8%，与未使用通道建模的Baseline方法相比，最高可提升3.2个百分点。

关键词: 目标检测, 上下文信息, 通道建模, 卷积神经网络, 特征增强

CLC Number:

TP391.41

ZHANG Yexing, CHEN Min, PAN Qiuyu. Object Detection Method Based on Feature Channel Modeling[J]. Computer Engineering, 2022, 48(7): 264-269,299.

张业星, 陈敏, 潘秋羽. 基于特征通道建模的目标检测方法[J]. 计算机工程, 2022, 48(7): 264-269,299.

/ / Recommend / Download Citations

URL: http://www.ecice06.com/EN/10.19678/j.issn.1000-3428.0061872

http://www.ecice06.com/EN/Y2022/V48/I7/264

Figures/Tables 9

References

[1] 程旭, 宋晨, 史金钢, 等.基于深度学习的通用目标检测研究综述[J].电子学报, 2021, 49(7):1428-1438. CHENG X, SONG C, SHI J G, et al.A survey of generic object detection methods based on deep learning[J].Acta Electronica Sinica, 2021, 49(7):1428-1438.(in Chinese)
[2] 周飞燕, 金林鹏, 董军.卷积神经网络研究综述[J].计算机学报, 2017, 40(6):1229-1251. ZHOU F Y, JIN L P, DONG J.Review of convolutional neural network[J].Chinese Journal of Computers, 2017, 40(6):1229-1251.(in Chinese)
[3] PRAMANIK A, SINGH H, DJEDDI C, et al.Region proposal and object detection using HoG-based CNN feature map[C]//Proceedings of International Conference on Data Analytics for Business and Industry:Way Towards a Sustainable Economy.Washington D.C., USA:IEEE Press, 2020:1-5.
[4] GUPTA S, KUMAR M, GARG A.Improved object recognition results using SIFT and ORB feature detector[J].Multimedia Tools and Applications, 2019, 78(23):34157-34171.
[5] GIRSHICK R.Fast R-CNN[C]//Proceedings of IEEE International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2015:1440-1448.
[6] REN S Q, HE K M, GIRSHICK R, et al.Faster R-CNN:towards real-time object detection with region proposal networks[C]//Proceedings of IEEE Transactions on Pattern Analysis and Machine Intelligence.Washington D.C., USA:IEEE Press, 2015:1137-1149.
[7] REDMON J, DIVVALA S, GIRSHICK R, et al.You only look once:unified, real-time object detection[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2016:779-788.
[8] REDMON J, FARHADI A.YOLO9000:better, faster, stronger[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:6517-6525.
[9] LIU W, ANGUELOV D, ERHAN D, et al.SSD:single shot multibox detector[C]//Proceedings of Conference on Computer Vision.Berlin, Germany:Springer, 2016:21-37.
[10] FU C Y, LIU W, RANGA A, et al.DSSD:decon-volutional single shot detector.[EB/OL].[2021-05-06].https://www.researchgate.net/publication/312759848.
[11] LIN T Y, DOLLÁR P, GIRSHICK R, et al.Feature pyramid networks for object detection[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:936-944.
[12] LI Z X, ZHOU F Q.FSSD:feature fusion single shot multibox detector[EB/OL].[2021-05-06].https://www.researchgate.net/publication/321511662_FSSD_Feature_Fusion_Single_Shot_Multibox_Detector.
[13] JEONG J, PARK H, KWAK N.Enhancement of SSD by concatenating feature maps for object detection[EB/OL].[2021-05-06].https://arxiv.org/abs/1705.09587.
[14] REDMON J, FARHADI A.YOLOv3:an incremental improvement[EB/OL].[2021-05-06].https://arxiv.org/abs/1804.02767.
[15] HE K M, ZHANG X Y, REN S Q, et al.Deep residual learning for image recognition[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2016:770-778.
[16] MHASKAR H N, POGGIO T.Deep vs.shallow networks:an approximation theory perspective[J].Analysis and Applications, 2016, 14(6):829-848.
[17] KETKAR N.Introduction to PyTorch[EB/OL].[2021-05-06].https://www.researchgate.net/publication/320201215_Introduction_to_PyTorch.
[18] SHRIVASTAVA A, GUPTA A, GIRSHICK R.Training region-based object detectors with online hard example mining[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2016:761-769.
[19] RUSSAKOVSKY O, DENG J, SU H, et al.ImageNet large scale visual recognition challenge[J].International Journal of Computer Vision, 2015, 115(3):211-252.
[20] EVERINGHAM M, ESLAMI S M A, GOOL L, et al.The pascal visual object classes challenge:a retrospective[J].International Journal of Computer Vision, 2015, 111(1):98-136.
[21] HOSANG J, BENENSON R, SCHIELE B.A convnet for non-maximum suppression[C]//Proceedings of Conference on Pattern Recognition.Berlin, Germany:Springer, 2016:192-204.
[22] MONTAVON G, ORR G B, MÜLLER K R.Neural networks:tricks of the trade:second edition[EB/OL].[2021-05-06].https://www.researchgate.net/publication/321620746.
[23] ROBERTSON S.A new interpretation of average precision[C]//Proceedings of the 31st Annual International ACM SIGIR Conference on Research and Development in Information Retrieval.New York, USA:ACM Press, 2008:689-690.

Please choose a citation manager

Content to export