《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (2): 574-583.DOI: 10.11772/j.issn.1001-9081.2021020324

• 多媒体计算与计算机仿真 • 上一篇    

基于分数阶网络和强化学习的图像实例分割模型

李学明1(), 吴国豪1, 周尚波1, 林晓然2, 谢洪斌3   

  1. 1.重庆大学 计算机学院, 重庆 400044
    2.河北经贸大学 信息技术学院, 石家庄 050061
    3.外生成矿与矿山环境重庆市重点实验室(重庆地质矿产研究院), 重庆 400042
  • 收稿日期:2021-03-04 修回日期:2021-04-29 接受日期:2021-04-30 发布日期:2021-07-29 出版日期:2022-02-10
  • 通讯作者: 李学明
  • 作者简介:李学明(1967—),男,重庆人,教授,博士,主要研究方向:深度学习、数据挖掘;
    吴国豪(1997—),男,山东菏泽人,硕士研究生,主要研究方向:强化学习、分数阶非线性系统;
    周尚波(1963—),男,广西宁明人,教授,博士,主要研究方向:视频图像信号处理、混沌控制理论;
    林晓然(1983—),女,河北石家庄人,讲师,博士,主要研究方向:图像信号处理、非线性系统、图像处理;
    谢洪斌(1985—),男,四川南部人,高级工程师,硕士,主要研究方向:遥感地质、遥感图像处理。
  • 基金资助:
    河北省高等学校科学技术研究项目(QN2019069);重庆市自然科学基金面上项目(cstc2019jcyj-msxmX0657)

Image instance segmentation model based on fractional-order network and reinforcement learning

Xueming LI1(), Guohao WU1, Shangbo ZHOU1, Xiaoran LIN2, Hongbin XIE3   

  1. 1.College of Computer Science,Chongqing University,Chongqing 400044,China
    2.School of Information Technology,Hebei University of Economics and Business,Shijiazhuang Hebei 050061,China
    3.Chongqing Key Laboratory of Exogenic Mineralization and Mine Environment (Chongqing Institute of Geology and Mineral Resources),Chongqing 400042,China
  • Received:2021-03-04 Revised:2021-04-29 Accepted:2021-04-30 Online:2021-07-29 Published:2022-02-10
  • Contact: Xueming LI
  • About author:LI Xueming, born in 1967, Ph. D., professor. His research interests include deep learning, data mining.
    WU Guohao, born in 1997, M. S. candidate. His research interests include reinforcement learning, fractional-order nonlinear system.
    ZHOU Shangbo, born in 1963, Ph. D., professor. His research interests include video and image signal processing, chaos control theory.
    LIN Xiaoran, born in 1983, Ph. D., lecturer. Her research interests include image signal processing, nonlinear system, image processing.
    XIE Hongbin, born in 1985, M. S., senior engineer. His research interests include remote sensing in geology, remote sensing image processing.
  • Supported by:
    Science and Technology Research Project of Higher Education of Hebei Province(QN2019069);Surface Program of Chongqing Natural Science Foundation(cstc2019jcyj-msxmX0657)

摘要:

针对目前的分数阶非线性模型图像特征提取能力不足导致分割精度较低的问题,提出一种基于分数阶网络和强化学习(RL)的图像实例分割模型,用来分割出图像中目标实例的高质量轮廓曲线。该模型共包含两层模块:1)第一层为二维分数阶非线性网络,主要采用混沌同步方法来获取图像中像素点的基础特征,并通过根据像素点间的相似性进行耦合连接的方式获取初步的图像分割结果;2)第二层通过RL思想将图像实例分割建立为一个马尔可夫决策过程(MDP),并利用建模过程中的动作-状态对、奖励函数和策略的设计来获取图像的区域结构和类别信息。最后将第一层获取到的像素特征和初步的图像分割结果与第二层获取到的区域结构和类别信息联合起来进行实例分割。在Pascal VOC2007 和Pascal VOC2012数据集上的实验结果表明,这种基于连续决策的图像实例分割模型与传统的分数阶模型相比,平均精度(AP)至少提升了15个百分点,不仅能够获取图像中目标物体的类别信息,而且进一步提升了对图像轮廓细节和细粒度信息的提取能力。

关键词: 强化学习, 分数阶网络, 混沌同步, 混沌吸引子, 马尔可夫决策过程, 像素-动作策略

Abstract:

Aiming at the low segmentation precision caused by the lack of image feature extraction ability of the existing fractional-order nonlinear models, an instance segmentation model based on fractional-order network and Reinforcement Learning (RL) was proposed to generate high-quality contour curves of target instances in the image. The model consists of two layers of modules: 1) the first layer was a two-dimensional fractional-order nonlinear network in which the chaotic synchronization method was mainly utilized to obtain the basic characteristics of the pixels in the image, and the preliminary segmentation result of the image was acquired through the coupling and connection according to the similarity among the pixels; 2) the second layer was to establish instance segmentation as a Markov Decision Process (MDP) based on the idea of RL, and the action-state pairs, reward functions and strategies during the modeling process were designed to extract the region structure and category information of the image. Finally, the pixel features and preliminary segmentation result of the image obtained from the first layer were combined with the region structure and category information obtained from the second layer for instance segmentation. Experimental results on datasets Pascal VOC2007 and Pascal VOC2012 show that compared with the existing fractional-order nonlinear models, the proposed model has the Average Precision (AP) improved by at least 15 percentage points, verifying that the sequential decision-based instance segmentation model not only can obtain the class information of the target objects in the image, but also further enhance the ability to extract contour details and fine-grained information of the image.

Key words: Reinforcement Learning (RL), fractional-order network, chaos synchronization, chaotic attractor, Markov Decision Process (MDP), pixel-action strategy

中图分类号: