属性一致的物体轮廓划分模型

孙劲光; 李桃; 董祥军

doi:10.11999/JEIT200741

属性一致的物体轮廓划分模型

doi: 10.11999/JEIT200741

孙劲光^{1, 2, ,},
李桃¹,
董祥军³

1.
辽宁工程技术大学葫芦岛 125105
2.
辽宁矿山安全数据工程技术研究中心葫芦岛 125105
3.
齐鲁工业大学(山东省科学院) 济南 250353

基金项目: 国家自然科学基金(61702241, 61602226)，国家重点研发计划(2018YFB1402902, 2018YFB1403303)

详细信息

作者简介:
孙劲光：女，1962年生，教授，研究方向为计算机图像视频处理与多媒体技术、计算机图形学与虚拟现实、数据科学与大数据计算

李桃：女，1986年生，博士生，研究方向为计算机图像视频处理与多媒体技术、数据科学与大数据计算

董祥军：男，1968年生，教授，研究方向为数据挖掘、人工智能和机器学习

通讯作者:
孙劲光　sunjinguang@lntu.edu.cn

中图分类号: TN911.73; TP399
计量
- 文章访问数: 557
- HTML全文浏览量: 354
- PDF下载量: 30
- 被引次数: 0
出版历程
- 收稿日期: 2020-08-24
- 修回日期: 2021-03-15
- 网络出版日期: 2021-03-25
- 刊出日期: 2021-10-18

Object Contour Partition Model with Consistent Properties

Jinguang SUN^{1, 2
, ,},
Tao LI¹,
xiangjun DONG³

1.
Liaoning University of Engineering and Technology, Huludao 125105, China
2.
Liaoning Mine Safely Data Engineering Technology Research Center, Huludao 125105, China
3.
Qilu University of Technology(Shandong Academy of Sciences), Jinan 250353, China

Funds: The National Natural Science Foundation of China(61702241, 61602226)，The National Key R&D Program of China (2018YFB1402902, 2018YFB1403303)

摘要

摘要: 该文提出一种基于全卷积深度残差网络、结合生成式对抗网络思想的基于属性一致的物体轮廓划分模型。采用物体轮廓划分网络作为生成器进行物体轮廓划分；该网络运用结构相似性作为区域划分的重构损失，从视觉系统的角度监督指导模型学习；使用全局和局部上下文判别网络作为双路判别器，对区域划分结果进行真伪判别的同时，结合对抗式损失提出一种联合损失用于监督模型的训练，使区域划分内容真实、自然且具有属性一致性。通过实例验证了该方法的实时性、有效性。
- 区域划分 /
- 矿石粒度分析 /
- 扩张卷积 /
- 跳跃连接 /
- 对抗式损失
Abstract: A new object contour partition model based on the fully convolutional network, combined with the idea of generative counter network and consistent attributes is proposed. Firstly, the image region partition network is used as a generator to divide the image region. Then the structural similarity is used as the reconstruction loss of regional division to supervise and guide model learning from the perspective of visual system. Finally, the global and local context discrimination networks are used as double-path similarity to supervise the reconstruction loss of regional division and guide model learning from the discriminators to distinguish the truth and falsity of the results of regional division, and a joint loss is proposed to train the supervision model in combination with the adversarial loss, so as to make the content of regional division true, natural and with attribute consistency. The instantaneity and effectiveness of the method are verified by living examples.
- Region division /
- Rock granularity analyze /
- Dilated convolution /
- Skip connection /
- Adversarial loss

HTML全文

图 1 属性一致的物体轮廓划分模型框架

下载: 全尺寸图片幻灯片

图 2 不同等级的样本制作图

下载: 全尺寸图片幻灯片

图 3 原图片

下载: 全尺寸图片幻灯片

图 4 传统方法的矿石轮廓划分图

下载: 全尺寸图片幻灯片

图 5 ReUnet大矿石轮廓划分图

下载: 全尺寸图片幻灯片

图 6 小矿石轮廓划分图

下载: 全尺寸图片幻灯片

图 7 矿石轮廓划分图

下载: 全尺寸图片幻灯片

表 1 物体区域划分网络结构

类别层	卷积核	池化	残差	膨胀	重塑	步长	输出
conv1	7×7	max	–	–	–	2	64×64×64
residual conv	3×3	–	4	–	–	2	8×8×512
dilated conv	3×3	–	–	3	–	2	8×8×512
reshape conv	1×1	–	–	–	4	1	256×256×256
conv2	3×3					2	256×256×256
conv3	1×1	–	–	–	–	1	256×256×1

下载: 导出CSV

表 2 全局上下文判别网络结构

类别层	卷积核	步长	输出
conv	5×5	2×2	64
conv	5×5	2×2	128
conv	5×5	2×2	256
conv	5×5	2×2	512
conv	5×5	2×2	512
conv	5×5	2×2	512
FC	–	–	1024

下载: 导出CSV

表 3 局部上下文判别网络结构

类别层	卷积核	步长	输出
conv	5×5	2×2	128
conv	5×5	2×2	256
conv	5×5	2×2	512
conv	5×5	2×2	512
conv	5×5	2×2	512
FC	–	–	1024

下载: 导出CSV

表 4 大矿石轮廓划分准确率

方法	识别总数	准确度		误判度		轮廓区域的准确率(%)
方法	识别总数	数量	准确率(%)	数量	误判率(%)	>90	80～90	70～80	60～70
传统方法^[16]	2370	1650	69.62	120	5.06	53.16	7.59	7.59	1.27
ReUnet	2370	2364	99.75	5	0.21	97.34	2.66	–	–

下载: 导出CSV

表 5 小矿石轮廓划分结果

方法	识别总数	准确度		误判度		轮廓区域的误识率(%)
方法	识别总数	数量	准确率(%)	数量	误判率(%)	小变大	大变小
传统方法^[16]	28812	13230	45.92	9114	68.89	6.67	62.22
ReUnet	28812	28518	98.98	294	1.02	–	1.02

下载: 导出CSV

参考文献(16)

[1]	SHELHAMER E, LONG J, and DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(4): 640–651. doi: 10.1109/TPAMI.2016.2572683
[2]	LIU Nian and HAN Junwei. DHSnet: Deep hierarchical saliency network for salient object detection[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 678–686. doi: 10.1109/CVPR.2016.80.
[3]	LI Guanbin and YU Yizhou. Deep contrast learning for salient object detection[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 478–487. doi: 10.1109/CVPR.2016.58.
[4]	XIE Saining and TU Zhuowen. Holistically-nested edge detection[C]. Proceedings of 2015 IEEE International Conference on Computer Vision, Santiago, Chile, 2015: 1395–1403. doi: 10.1109/ICCV.2015.164.
[5]	ZHAO Hengshuang, SHI Jianping, QI Xiaojuan, et al. Pyramid scene parsing network[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 6230–6239. doi: 10.1109/cvpr.2017.660.
[6]	BADRINARAYANAN V, KENDALL A, and CIPOLLA R. SegNet: A deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12): 2481–2495. doi: 10.1109/TPAMI.2016.2644615
[7]	余春艳, 徐小丹, 钟诗俊. 融合去卷积与跳跃嵌套结构的显著性区域检测[J]. 计算机辅助设计与图形学学报, 2018, 30(11): 2150–2158. YU Chunyan, XU Xiaodan, and ZHONG Shijun. Saliency region detection based on deconvolutional and skip nested module[J]. Journal of Computer-Aided Design &Computer Graphics, 2018, 30(11): 2150–2158.
[8]	张冬明, 靳国庆, 代锋, 等. 基于深度融合的显著性目标检测算法[J]. 计算机学报, 2019, 42(9): 2076–2086. doi: 10.11897/SP.J.1016.2019.02076 ZHANG Dongming, JIN Guoqing, DAI Feng, et al. Salient object detection based on deep fusion of hand-crafted features[J]. Chinese Journal of Computers, 2019, 42(9): 2076–2086. doi: 10.11897/SP.J.1016.2019.02076
[9]	HOU Qibin, CHENG Mingming, HU Xiaowei, et al. Deeply supervised salient object detection with short connections[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(4): 815–828. doi: 10.1109/TPAMI.2018.2815688
[10]	纪超, 黄新波, 曹雯, 等. 结合深度学习和全局-局部特征的图像显著区域计算[J]. 计算机辅助设计与图形学学报, 2019, 31(10): 1838–1846. doi: 10.3724/SP.J.1089.2019.17544 JI Chao, HUANG Xinbo, CAO Wen, et al. Fusion of deep learning and global-local features of the image salient region calculation[J]. Journal of Computer-Aided Design &Computer Graphics, 2019, 31(10): 1838–1846. doi: 10.3724/SP.J.1089.2019.17544
[11]	WU Huikai, ZHANG Junge, HUANG Kaiqi, et al. FastFCN: Rethinking dilated convolution in the backbone for semantic segmentation[J]. arXiv: 1903.11816, 2019.
[12]	郭辉, 芮兰兰, 高志鹏. 车辆边缘网络中基于多参数MDP模型的动态服务迁移策略[J]. 通信学报, 2020, 41(1): 1–14. doi: 10.11959/j.issn.1000-436x.2020012 GUO Hui, RUI Lanlan, and GAO Zhipeng. Dynamic service migration strategy based on MDP model with multiple parameter in vehicular edge network[J]. Journal on Communications, 2020, 41(1): 1–14. doi: 10.11959/j.issn.1000-436x.2020012
[13]	范九伦, 雷博. 倒数粗糙熵图像阈值化分割算法[J]. 电子与信息学报, 2020, 42(1): 214–221. doi: 10.11999/JEIT190559 FAN Jiulun and LEI Bo. Image thresholding segmentation method based on reciprocal rough entropy[J]. Journal of Electronics &Information Technology, 2020, 42(1): 214–221. doi: 10.11999/JEIT190559
[14]	廖苗, 李阳, 赵于前, 等. 一种新的图像超像素分割方法[J]. 电子与信息学报, 2020, 42(2): 364–370. doi: 10.11999/JEIT190111 LIAO Miao, LI Yang, ZHAO Yuqian, et al. A new method for image superpixel segmentation[J]. Journal of Electronics &Information Technology, 2020, 42(2): 364–370. doi: 10.11999/JEIT190111
[15]	LI He, LI Gang, WANG Xuehu, et al. Edge detection of heterogeneity in transmission images based on frame accumulation and multiband information fusion[J]. Chemometrics and Intelligent Laboratory Systems, 2020, 204: 104117. doi: 10.1016/j.chemolab.2020.104117
[16]	刘健庄, 栗文青. 灰度图象的二维Otsu自动阈值分割法[J]. 自动化学报, 1993, 19(1): 101–105. doi: 10.16383/j.aas.1993.01.015 LIU Jianzhuang and LI Wenqing. The automatic thresholding of gray-level pictures via two-dimensional Otsu method[J]. Acta Automatica Sinica, 1993, 19(1): 101–105. doi: 10.16383/j.aas.1993.01.015