基于图小波的几何深度学习及其在3D点云中的应用研究
结题报告
批准号:
61966007
项目类别:
地区科学基金项目
资助金额:
38.0 万元
负责人:
张文辉
依托单位:
学科分类:
机器学习
结题年份:
2023
批准年份:
2019
项目状态:
已结题
项目参与者:
林基明
国基评审专家1V1指导 中标率高出同行96.8%
结合最新热点,提供专业选题建议
深度指导申报书撰写,确保创新可行
指导项目中标800+,快速提高中标率
客服二维码
微信扫码咨询
中文摘要
本项目着眼于动态3D点云环境中的目标识别、视频预测,以帮助智能体实现准确、实时的环境感知与智能决策。拟基于图信号处理技术,将几何深度学习方法与图小波技术相融合,以提升3D点云处理性能。首先将具有时频局部化的图小波作为几何深度学习网络的卷积核形成新的网络体系,研究图小波与几何深度学习网络融合的机制,揭示图小波特性与几何深度学习网络特性之间的关系,提升单帧点云目标识别的准确度;其次研究匹配于动态输入信号的自适应小波生成方法,将自适应小波与能处理含时任务的深度学习网络相结合生成新的网络结构体系,以提升3D视频预测序列的准确度与实时性;在此基础上,将几何深度学习方法与强化学习方法融合设计有效的深度强化学习算法实现智能体的智能决策。本研究具有重要的工程应用价值和深远的科学意义。
英文摘要
This Project focus on target recognition and video prediction in dynamic 3D point cloud environment to help agents accurately and real-time perceive and make intelligent decisions. It is proposed to root in graph signal processing, to integrate geometric deep learning methods with graph wavelets for the purpose of improving the processing performance of 3D point clouds. Firstly, graph wavelets with time-frequency localization is used as the convolution kernel of geometric deep learning network to form a new network system. The fusion mechanism of graph wavelets and geometric deep learning network is studied to reveal the relationship between graph wavelets and geometric deep learning network characteristics and improve the accuracy of single frame point clouds recognition. Secondly, an adaptive wavelet generation method matching with dynamic input signals is studied, and a new network architecture is generated by integrating the adaptive wavelet with the deep learning network capable of processing time-dependent tasks, so as to improve the accuracy and real-time performance of 3D video prediction sequences. Finally, the geometric deep learning method and reinforcement learning method are combined to design an effective deep reinforcement learning algorithm to realize intelligent decision making of agents. The research has important values on engineering application as well as significant science meanings.
期刊论文列表
专著列表
科研奖励列表
会议论文列表
专利列表
DOI:--
发表时间:2021
期刊:计算机应用与软件
影响因子:--
作者:薛磊;农丽萍;张文辉;林基明;王俊义
通讯作者:王俊义
DOI:10.1109/twc.2021.3062616
发表时间:2021-08-01
期刊:IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS
影响因子:10.4
作者:Peng, Jie;Qiu, Hongbing;Wang, Junyi
通讯作者:Wang, Junyi
DOI:--
发表时间:2023
期刊:桂林电子科技大学学报
影响因子:--
作者:甘 萍;林基明;农丽萍;王俊义
通讯作者:王俊义
DOI:10.3969/j.issn.1000-8349.2020.04.05
发表时间:2020
期刊:天文学进展
影响因子:--
作者:于昂;劳保强;王俊义;安涛
通讯作者:安涛
DOI:10.1142/s0219691320500848
发表时间:2020
期刊:International Journal of Wavelets, Multiresolution and Information Processing
影响因子:--
作者:Wenhui Zhang;Chenyu Wang;Wenjie Lin;Jiming Lin
通讯作者:Jiming Lin
国内基金
海外基金