[Oral Presentation]基于深度学习的颗粒大小和形状表征方法

Organized By

Hohai University

Monash University

Co-organized by

Southeast University

Jiangxi University of Science and Technology

Northeast University

DUT-BSU Scientific Research and Innovation Center

Suzhou University of Science and Technology

State Key Laboratory of Structural Analysis, Optimization and CAE Software for Industrial Equipment

Sponsoring agency

KELI SDEM Software (Dalian) Co., Ltd.

Sugon Smart Computing Information Technology Co., Ltd.

Beijing Paratera Technology Co., Ltd.

Simpop Information Technology Co., Ltd.

Important Dates

Conference Announcement

December 2023, release First Round Announcement
Abstract Submission Deadline
March 31^th, 2024
Paper Submission Deadline
April 30^th, 2024
Registration
From January 1^st, 2024
Conference Schedule
May 31^st (Friday), 2024 On-Site Registration
June 1^st to 3^rd, 2024 Conference

基于深度学习的颗粒大小和形状表征方法

ID：80 Submission ID：232 View Protection：ATTENDEE Updated Time：2024-04-09 22:37:53 Hits：583 Oral Presentation

Start Time：Pending （Asia/Shanghai）

Duration：Pending

Session：[No Session] » [No Session Block]

No files

Abstract

粗粒土的力学性质与其颗粒形状和粒径密切相关。传统方法通过筛分试验确定颗粒粒径，并采用与标准图表对比定性描述颗粒形状。然而，传统方法存在许多不足和限制，难以适用于粒径过大和难以采样等特殊情况。因此，本文基于深度学习提出了一种针对图像识别的粗粒土粒径和形状量化方法。该方法采用实例分割模型SOLOv2，结合可变形卷积网络对图像中堆叠的粗粒土进行准确分割。然后，依据计算几何方法对分割颗粒的形状和粒径进行量化分析，并将量化结果与实际结果进行对比。结果表明，本文提出的方法与传统方法的测量结果基本一致，验证了提出方法的准确性。另外，采用该方法对广西武鸣河和火星上的粗粒土进行分析测试，结果符合标准图表的视觉预期，这为准确量化粗粒土的颗粒形状和粒径提供一种新途径。

Keywords

粗粒土，颗粒形状，粒径，图像分割，深度学习

Speaker

刘子杨

广西大学

Submission Author

刘子杨广西大学

龚健广西大学

Sponsored By

Hohai University

Panel of Computational Mechanics on Granular Materials, Working Party of Computational Mechanics, Chinese Society of Theoretical and Applied Mechanics

Jiangsu Society of Theoretical and Applied Mechanics