基于改进密集网络与二次回归的小目标检测算法

doi:10.19678/j.issn.1000-3428.0057595

摘要/Abstract

摘要： 基于单激发探测器（SSD）的小目标检测算法实时性较差且检测精度较低。为提高小目标检测精度和鲁棒性，提出一种结合改进密集网络和二次回归的小目标检测算法。将SSD算法中骨干网络由VGG16替换为特征提取能力更强且速度更快的DenseNet，利用基于区域候选的检测算法中默认框由粗到细筛选的回归思想设计串级SSD网络结构，在区分目标和背景后进行常规目标分类和位置回归，以获取精确的默认框信息并达到小目标检测中正负样本比例均衡。在此基础上，使用特征图尺度变换方法在不增加参数量情况下完成特征图融合，同时通过K-means聚类方法得到默认框的最佳长宽比并重新设置其尺寸。实验结果表明，该算法的检测平均精度均值在PASCAL VOC2007公共数据集和自制航拍小目标数据集上分别为82.3%和87.6%，较改进前SSD算法分别提升5.1个百分点和9.5个百分点，检测速度达到58 frames/s，可有效实现小目标的实时性检测。

关键词: 单激发探测器检测算法, 深度学习, 小目标检测, 密集网络, 二次回归

Abstract: The small object detection algorithms based on Single Shot Detector(SSD) have poor real-time performance and low detection accuracy.In order to improve the accuracy and robustness of small target detection,this paper proposes a small object detection algorithm combining the improved dense network and quadratic regression.The backbone network in the SSD algorithm,VGG16,is replaced with DenseNet that has stronger feature extraction ability and higher speed.The structure of serial SSD network is designed by using the regression idea of coarse-to-fine screening of default box in the detection algorithm based on region candidates.The object and background are distinguished first,and then the conventional object classification and position regression are carried out to obtain accurate default fox information and achieve the balance of the proportion of positive and negative samples in small object detection.On this basis,the feature graph fusion is completed without increasing the number of parameters by using the scale transformation method of feature graph.At the same time,the best aspect ratio of the default box is obtained by using the K-means clustering method,and its size is reset.Experimental results show that the detection mean Average Precision(mAP) of the algorithm is 82.3 percent and 87.6 percent on PASCAL VOC2007 public dataset and self-made aerial photographing small target dataset,respectively,increased by 5.1 and 9.5 percentage points compared with that of the previous SSD algorithm.The detection speed of the algorithm reaches 58 frames/s,which demonstrates that it can effectively realize the real-time detection of small objects.

Key words: Single Shot Detector(SSD) detection algorithm, deep learning, small object detection, dense network, quadratic regression

中图分类号:

TP391.4

奚琦, 张正道, 彭力. 基于改进密集网络与二次回归的小目标检测算法[J]. 计算机工程, 2021, 47(4): 241-247,255.

XI Qi, ZHANG Zhengdao, PENG Li. Small Object Detection Algorithm Based on Improved Dense Network and Quadratic Regression[J]. Computer Engineering, 2021, 47(4): 241-247,255.

http://www.ecice06.com/CN/Y2021/V47/I4/241

图/表 12

20210425171415

20210425171417

20210425171421

20210425171424

20210425171427

20210425171430

20210425171433

20210425171436

20210425171438

20210425171441

20210425171444

20210425171446

参考文献

[1] ZONG Yanfen.Research on the quality detection system and method of micro parts based on computer vision[D].Harbin:Harbin University of Science and Technology,2017.(in Chinese)宗艳芬.基于计算机视觉的微小零件质量检测系统和方法研究[D].哈尔滨:哈尔滨理工大学,2017.
[2] CHEN Yun.Medical image detection algorithm based on deep learning[D].Beijing:Beijing University of Posts and Telecommunications,2019.(in Chinese)陈云.基于深度学习的医学影像检测算法[D].北京:北京邮电大学,2019.
[3] SUN Xinde,FANG Guizhen,LI Lingling,et al.Infrared small target detection via combination of morphological filtering and MRF model[J].Computer Engineering,2012,38(14):153-156.(in Chinese)孙新德,方桂珍,李玲玲,等.联合形态滤波和MRF模型的红外小目标检测[J].计算机工程,2012,38(14):153-156.
[4] BELL S,ZITNICK C L,BALA K,et al.Inside-outside net:detecting objects in context with skip pooling and recurrent neural networks[C]//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C.,USA:IEEE Press,2015:2874-2883.
[5] MAENPAA T,PIETIKAINEN M.Image analysis with local binary patterns[C]//Proceedings of 2005 Scandinavian Conference on Image Analysis.Berlin,Germany:Springer,2005:115-118.
[6] PANNING A,AL-HAMADI A K,NIESE R,et al.Facial expression recognition based on Haar-like feature detection[J].Pattern Recognition and Image Analysis,2008,18(3):447-452.
[7] DALAL N,TRIGGS B.Histograms of oriented gradients for human detection[C]//Proceedings of 2005 IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C.,USA:IEEE Press,2005:886-893.
[8] LOWE D G.Distinctive image features from scale-invariant keypoints[J].International Journal of Computer Vision,2004,60(2):91-110.
[9] GIRSHICK R.Fast R-CNN[C]//Proceedings of 2015 IEEE International Conference on Computer Vision.Washington D.C.,USA:IEEE Press,2015:1440-1448.
[10] REN S,HE K,GIRSHICK R,et al.Faster R-CNN:towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,39(6):91-99.
[11] REDMON J,DIVVALA S,GIRSHICK R,et al.You only look once:unified,real-time object detection[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C.,USA:IEEE Press,2016:779-788.
[12] LIU W,ANGUELOV D,ERHAN D,et al.SSD:single shot multibox detector[C]//Proceedings of 2016 European Conference on Computer Vision.Berlin,Germany:Springer,2016:21-37.
[13] FU C Y,LIU W,RANGA A,et al.DSSD:deconvolutional single shot detector[EB/OL].[2020-01-20].https://arxiv.org/abs/1701.06659.
[14] SIMONYAN K,ZISSERMAN A.Very deep convolutional networks for large-scale image recognition[EB/OL].[2020-01-20].https://arxiv.org/abs/1409.1556.
[15] HE Kaiming,ZHANG Xiangyu,REN Shaoqing,et al.Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C.,USA:IEEE Press,2016:770-778.
[16] JEONG J,PARK H,KWAK N.Enhancement of SSD by concatenating feature maps for object detection[EB/OL].[2020-01-20].https://arxiv.org/abs/1705.09587.
[17] LI Zuoxin,ZHOU Fuqiang.FSSD:feature fusion single shot multibox detector[EB/OL].[2020-01-20].https://arxiv.org/abs/1712.00960.
[18] CUI Lisha,MA Rui,LÜ Pei,et al.MDSSD:multi-scale de-convolutional single shot detector for small objects[EB/OL].[2020-01-20].https://arxiv.org/abs/1805.07009.
[19] LIU Songtao,HUANG Di,WANG Yunhong.Receptive field block net for accurate and fast object detection[C]//Proceedings of 2018 European Conference on Computer Vision,Berlin,Germany:Springer,2018:31-39.
[20] HUANG G,LIU Z,MAATEN L V D,et al.Densely connected convolutional networks[C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C.,USA:IEEE Press,2017:2261-2269.
[21] SHRIVASTAVA A,GUPTA A,GIRSHICK R.Training region-based object detectors with online hard example mining[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C.,USA:IEEE Press,2016:761-769.
[22] LIN T Y,GOYAL P,GIRSHICK R,et al.Focal loss for dense object detection[C]//Proceedings of 2017 IEEE International Conference on Computer Vision.Washington D.C.,USA:IEEE Press,2017:2999-3007.

选择文件类型/文献管理软件名称

选择包含的内容