残差网络下基于困难样本挖掘的目标检测

张超; 陈莹

doi:doi:10.3788/lop55.101003

激光与光电子学进展, 2018, 55 (10): 101003, 网络出版: 2018-10-14

残差网络下基于困难样本挖掘的目标检测下载： 902次

Object Detection Based on Hard Examples Mining Using Residual Network

论文大纲

张超 ^*陈莹

作者单位

江南大学轻工过程先进控制教育部重点实验室, 江苏无锡 214122

AI 词云图 AI语音精读 AI语音超短摘要

注：本部分内容由 AI 自动生成，请您知悉。

摘要

为了提高图像目标的检测精度, 提出一种在残差网络下设计基于困难样本挖掘的目标检测算法。首先阐述基于深度学习的目标检测算法, 即超快速区域卷积神经网络(Faster R-CNN)的工作原理, 分析该算法存在的不足与改进方式。在Faster R-CNN的基础上, 为了使模型能提取更有效的深度卷积特征, 选取网络更深的残差网络替换原始的ZF或VGG网络。为了使学习到的网络模型有更强的泛化能力, 在网络训练过程中, 利用困难样本更新网络参数, 使网络训练更充分。在Pascal VOC2007、Pascal VOC2007+Pascal VOC2012和BIT这三个数据集中进行训练和测试, 实验结果显示, 融合了两种方法的Faster R-CNN在这三个数据集上的检测精度分别提升了3.5%、7.1%、6.4%, 提升效果明显。

Abstract

In order to detect objects more accurately in images, an object detection algorithm based on hard example mining and residual network is proposed, which takes faster regional convolutional neural network (Faster R-CNN) as a benchmark. The working principle of Faster R-CNN is described based on deep learning, and the shortcomings and improvement methods of the algorithm are analyzed. Specifically, a deeper residual network is adopted to replace the original ZF or VGG network to extract more effective deep convolution features. In order to enhance the generalization ability of the learning network model, the network parameters are updated with hard examples during training. The experimental results on Pascal VOC2007, Pascal VOC2007+Pascal VOC2012 and BIT show that compared with Faster R-CNN, the proposed method improves detection accuracy by 3.5%, 7.1%, 6.4%, respectively, on the above three datasets.

参考文献

[1] Aggarwal J K, Ryoo M S. Human activity analysis: a review［J］. ACM Computing Surveys, 2011, 43(3): 16.

[2] Datta R, Joshi D, Li J, et al. Image retrieval: ideas, influences, and trends of the new age［J］. ACM Computing Surveys, 2008, 40(2): 5.

[3] Krüger V, Kragic D, Ude A, et al. The meaning of action: a review on action recognition and mapping［J］. Advanced Robotics, 2007, 21(13): 1473-1501.

[4] Palmese M, Trucco A. From 3D sonar images to augmented reality models for objects buried on the seafloor［J］. IEEE Transactions on Instrumentation and Measurement, 2008, 57(4): 820-828.

[5] 陈银, 任侃, 顾国华, 等. 基于改进的单高斯背景模型运动目标检测算法［J］. 中国激光, 2014, 41(11): 1109002.

Chen Y,Ren K, Gu G H, et al. Moving object detection based on improved single gaussian background model［J］. Chinese Journal of Lasers, 2014, 41(11): 1109002.

[6] 刘德连, 张建奇, 何国经. 背景高斯化的遥感图像目标检测［J］. 光学学报, 2007, 27(4): 638-642.

Liu D L, Zhang J Q, He G J. Target detection for remote sensing image based on gaussian transformation of background［J］. Acta Optica Sinica, 2007, 27(4): 638-642.

[7] 卢清华, 吴志伟, 范彦斌, 等. 基于混合高斯模型的运动车辆检测方法［J］. 光电子·激光, 2013, 24(4): 751-757.

Lu Q H, Wu Z W, Fan Y B, et al. An improved mobile vehicle detection method based on Gaussian mixture model［J］. Journal of Optoelectronics·Laser, 2013, 24(4): 751-757.

[8] Felzenszwalb P F, Girshick R B, McAllester D, et al. Object detection with discriminatively trained part-based models［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 32(9): 1627-1645.

[9] Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks［C］. Advances in Neural Information Processing Systems, 2012: 1097-1105.

[10] 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.

[11] Girshick R. Fast R-CNN［C］∥Proceedings of the IEEE International Conference on Computer Vision, 2015: 1440-1448.

[12] Ren S Q, He K M, 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, 2017, 39(6): 1137-1149.

[13] He K, Gkioxari G, Dollár P, et al. Mask R-CNN［C］∥Proceedings of the IEEE International Conference on Computer Vision, 2017: 2980-2988.

[14] 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, 2016: 779-788.

[15] Liu W, Anguelov D, Erhan D, et al. SSD: Single shot multibox detector［C］. European Conference on Computer Vision, 2016: 21-37.

[16] 杨名, 阮雅端, 陈林凯, 等. 甚高速区域卷积神经网络的船舶视频检测方法［J］. 北京邮电大学学报, 2017, 40(S1): 130-134.

Yang M, Ruan Y D, Chen L K, et al. New video recognition algorithms for inland river ships based on faster R-CNN［J］. Journal of Beijing University of Posts and Telecommunications, 2017, 40(S1): 130-134.

[17] 曹诗雨, 刘跃虎, 李辛昭. 基于Fast R-CNN的车辆目标检测［J］. 中国图象图形学报, 2017, 22(5): 671-677.

Cao S Y, Liu Y H, Li X Z. Vehicle detection method based on fast R-CNN［J］. Journal of Image and Graphics, 2017, 22(5): 671-677.

[18] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition［C］∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.

[19] Zeiler M D, Fergus R. Visualizing and understanding convolutional networks［C］. European Conference on Computer Vision, 2014: 818-833.

[20] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition［J］. arXiv: 1409.1556, 2014.

[21] Ouyang W, Wang X, Zhang C, et al. Factors in finetuning deep model for object detection with long-tail distribution［C］∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 864-873.

[22] 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, 2016: 761-769.

[23] Dong Z, Jia Y. Vehicle type classification using distributions of structural and appearance-based features［C］∥Proceedings of the IEEE International Conference on Image Processing, 2013: 4321-4324.

张超, 陈莹. 残差网络下基于困难样本挖掘的目标检测[J]. 激光与光电子学进展, 2018, 55(10): 101003. Zhang Chao, Chen Ying. Object Detection Based on Hard Examples Mining Using Residual Network[J]. Laser & Optoelectronics Progress, 2018, 55(10): 101003.

残差网络下基于困难样本挖掘的目标检测下载： 902次

关于本站 Cookie 的使用提示

全站搜索

残差网络下基于困难样本挖掘的目标检测 下载： 902次

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索

残差网络下基于困难样本挖掘的目标检测下载： 902次