基于强化学习的图像不确定性目标域提取仿真  

Extracting Simulation Based on Enhanced Learning Image Uncertainty Target Domain

作　　者：代丽娜 裴冬菊[2] 郑冬花 叶丽珠 DAI Li-na;PEI Dong-ju;ZHENG Dong-hua;YE Li-zhu(School of Information Technology&Engineering,Guangzhou College of Commerce,Guangzhou Guangdong 511363,China;School of Computer and Information Engineering,Jiangxi Agricultural University,Nanchang Jiangxi 330045,China)

机构地区：[1]广州商学院信息技术与工程学院,广东广州511363 [2]江西农业大学计算机与信息工程学院,江西南昌330045

出　　处：《计算机仿真》2024年第3期145-148,378,共5页Computer Simulation

基　　金：江西省教育厅科技计划项目(GJJ210433);广东省高等教育学会“十四五”规划2021年度高等教育研究课题(21GYB08);广东省2022年本科高校教学质量与教学改革工程高等教育教学改革项目(2022SJJXGG991)。

摘　　要：相比于输入图像的整体像素点,不确定性目标所占像素点偏少,且由于物体朝向不同,对应特征也不同,导致图像不确定性目标提取的难度较大。提出一种新的基于强化学习的图像不确定性目标域提取方法。将图像输入加权双边滤波器中,将其分为高频区域和低频区域,在Curvelet变换的基础上保留图像细节信息。消除图像噪声,并采用限邻域EMD方法增强图像质量。建立自适应性模型,将预处理后的图像输入至上述模型中,通过混沌同步方法在分数阶非线性网络中提取图像像素点特征,以此实现图像分割。应用强化学习建立图像不确定性目标域提取的马尔科夫决策,获取图像的类别信息和区域结构,图像不确定性目标域的提取。仿真结果表明,研究方法能够精准提取图像像素点特征,查全率和查准率高于90%,J指数以及F检验值的平均值可达0.8以上。Compared with the overall pixels of the input image,the number of pixels occupied by uncertain targets is relatively small.Due to the different orientations of objects,their features are different as well,leading to the difficulty of extracting uncertain targets in images.Based on reinforcement learning,a new method was proposed to extract the uncertain target domains in the image.Firstly,the image was input into a weighted bilateral filter,and then it was divided into high-frequency region and low-frequency region.Based on the curvelet transform,image details were retained.Moreover,the image noise was eliminated,and then the image quality was enhanced by Neighborhood Limited EMD.Furthermore,an adaptive model was built,and then the preprocessed image was input into the model.Meanwhile,the pixel features were extracted from the fractional nonlinear network by the chaotic synchronization method,so that image segmentation could be achieved.Finally,the reinforcement learning method was adopted to establish a Markov decision for extracting the uncertain target domain,thus obtaining the category information and regional structure and extracting the uncertain target domain of the image.Simulation results show that the proposed method can extract image pixel features accurately,with recall and precision higher than 90%.The mean value of index J and test value F can be more than O0.8.

关 键 词：强化学习 加权双边滤波器 混沌同步方法 马尔科夫决策过程 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]