基于多尺度联合权重分配的目标检测算法

doi:10.3778/j.issn.1002-8331.2109-0438

摘要/Abstract

摘要： 针对SSD（single shot multibox detector）算法在复杂交通场景下对交通标志小目标识别效果不佳的问题，提出一种基于多尺度特征互补和重点特征信息挖掘（multi-scale feature complementary fusion and key feature information mining SSD，MK-SSD）的目标检测改进算法。利用跨阶段局部网络设计多尺度特征互补模块，同时构建多路径特征融合网络，有效提升浅层网络对小目标的特征提取能力。设计联合权重分配模块，将感知域与重点信息挖掘相结合，更高效地利用重点特征信息并抑制对非重点信息的关注度。利用轻量化残差块对预测网络进行改进，提升目标检测能力。经实验分析，改进后的算法在自制交通标志数据集上平均准确率达到89.64%，在保证实时性的同时，相较于YOLO系列和SSD系列算法拥有更高的检测精度，能检测出大部分SSD网络漏检的小目标。

关键词: 小目标检测, 特征互补, 多路径特征融合, 权重分配, 残差块

Abstract: For the problem of SSD（single shot multibox detector） identification in complex traffic environments, this paper proposes a target detection improvement algorithm based on multi-scale feature complementarity and combined weight distribution（multi-scale feature complementary fusion and key feature information mining SSD, MK-SSD）. The proposed algorithm first designs multi-scale feature complementary modules using cross stage partial network and builds multi-path feature fusion networks to effectively improve the feature extraction capability of shallow networks for small targets. Secondly, the combined weight distribution module is designed to combine the perceptual domain with the key information mining to more efficiently use the key feature information and suppress the attention of the non-key information. Finally, the prediction network is improved using lightweight residual blocks to improve the target detection capability. After experimental analysis, the average accuracy of the improved algorithm reaches 89.64% on the homemade traffic sign dataset. While ensuring real-time, it has higher detection accuracy compared with YOLO series and SSD series algorithms, and can detect small targets of most SSD network leakage detection.

Key words: small object detection, feature complementarity, multi-path feature fusion, weight distribution, residual block

崔静雯, 马杰, 张宇. 基于多尺度联合权重分配的目标检测算法[J]. 计算机工程与应用, 2022, 58(17): 101-110.

CUI Jingwen, MA Jie, ZHANG Yu. Target Detection Algorithm Based on Multi-Scale Combined Weight Distribution[J]. Computer Engineering and Applications, 2022, 58(17): 101-110.

参考文献

[1] 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，2014：580-587.
[2] REN S Q，HE K，GIRSHICK R，at al.Faster R-CNN：towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2017，39（6）：1137-1149.
[3] EDMON J，DIVVALA S，GIRSHICK R，et al.You only look once：unified，real-time object detection[C]//IEEE Conference on Computer Vision and Pattern Recognition（CVPR），Las Vegas，NV，USA，2016：779-788.
[4] REDMON J，FARHADI A.YOLO9000：better，faster，stronger[C]//IEEE Conference on Computer Vision and Pattern Recognition（CVPR），2017：6517-6525.
[5] REDMON J，FARHADI A.YOLOv3：an incremental improvement[C]//Proceedings of 2018 IEEE Conference on Computer Vision and Pattern Recognition（CVPR）.Washington DC，USA：IEEE Press，2018：1-6.
[6] BOCHKOVSKIY A，WANG C Y，LIAO H.YOLOv4：optimalspeed and accuracy of object detection[J].arXiv：2004.10934，2020.
[7] LIU W，ANGUELOV D，ERHAN D，et al.SSD：single shot multibox detector[C]//Computer Vision-ECCV 2016.Cham：Springer，2016：21-37.
[8] 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，2017：2117-2125.
[9] LI Z，ZHOU F.FSSD：feature fusion single shot multibox detector[J].arXiv：1712，00960，2017.
[10] FU C，LIU W，RANGA A，et al.DSSD：deconvolutional single shot detector[J].arXiv：1712.00960，2017.
[11] 李明山，韩清鹏，张天宇，等.改进SSD的安全帽检测方法[J].计算机工程与应用，2021，57（8）：192-197.
LI M S，HAN Q P，ZHANG T Y，et al.Safety helmet detection method of improved SSD[J].Computer Engineering and Applications，2021，57（8）：192-197.
[12] 刘峰，郭猛，王向军.基于跨尺度融合的卷积神经网络小目标检测[J].激光与光电子学进展，2021，58（6）：205-213.
LIU F，GUO M，WANG X J.Small target detection based on cross-scale fusion convolution neural network[J].Laser & Optoelectronics Progress，2021，58（6）：205-213.
[13] ZHANG J，HUANG M，JIN X，et al.A real-time Chinese traffic sign detection algorithm based on modified YOLOv2[J].Algorithms，2017，10（4）：127.
[14] SHAN H，ZHU W.A small traffic sign detection algorithm based on modified SSD[J].IOP Conference Series：Materials Science and Engineering，2019，646（1）：012006.
[15] WANG C Y，LIAO H Y M，WU Y H，et al.CSPNet：a new backbone that can enhance learning capability of CNN[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition（CVPR），2020：390-391.
[16] WU Z，SHEN C，VAN DEN HENGEL A.Wider or deeper：revisiting the resnet model for visual recognition[J].Pattern Recognition，2019，90：119-133.
[17] 王冬丽，廖春江，牟金震，等.基于特征融合的SSD视觉小目标检测[J].计算机工程与应用，2020，56（16）：31-36.
WANG D L，LIAO C J，MU J Z，et al.SSD visual small target detection based on feature fusion[J].Computer Engineering and Applications，2020，56（16）：31-36.
[18] ANTHIMOPOULOS M，CHRISTODOULIDIS S，EBNER L，et al.Semantic segmentation of pathologicallung tissue with dilated fully convolutional networks[J].IEEE Journal of Biomedical and Health Informatics，2019，23（2）：714-722.
[19] LI B，DAI Y，HE M.Monocular depth estimation with hierarchical fusion of dilated cnns and soft-weighted-sum inference[J].Pattern Recognition，2020，83：328-339.
[20] WANG Q，WU B，ZHU P，et al.ECA-Net：efficient channel attention for deep convolutional neural networks[C]//CVF Conference on Computer Vision and Pattern Recognition（CVPR），2020：1571-1580.
[21] HU J，SHEN L，SUN G.Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition（CVPR），Seattle，USA，June 13-19，2020.Piscataway：IEEE，2020-2023.
[22] WOO S，PARK J，LEE J Y，et al.CBAM：convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision，2018：3-19.
[23] SRINIVASU P N，SIVASAI J G，IJAZ M F，et al.Classification of skin disease using deep learning neural networks with MobileNet V2 and LSTM[J].Sensors，2021，21（8）：2852.