Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Civil and Environmental Engineering | en_US |
| dc.contributor.advisor | Yin, Zhen-yu (CEE) | en_US |
| dc.creator | Zhang, Haoran | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/13020 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Intelligent identification, reconstruction, and geometric evaluation of granular soils based on deep learning techniques | en_US |
| dcterms.abstract | Granular soils are commonly used as crucial construction materials in geotechnical engineering. The characteristics of morphology and grain size of the granular soils have an important impact on their mechanical properties. In the field inspection of onsite granular particles, the evaluation of such characteristics is required to maintain well mechanical properties in densely packed conditions. The conventional evaluation methods focus on the measurement of an individual particle in the laboratory, which is low-efficiency and infeasible in field applications. Recently, computer vision-based methods that can rapidly and efficiently recognize particles from numerous images have been developed for particle analysis. However, it is still challenging to deal with the identification and reconstruction of granular particles in 2D and 3D space for the densely packed granular soils in field inspection due to occlusions. Regarding the background, this thesis comprehensively proposes a strategy, two frameworks, and a model for the segmentation and completion of the granular soils ranging from 1mm to 15cm, and evaluates the morphology characteristics based on the segmented and completed results. | en_US |
| dcterms.abstract | Extracting the outlines of particles has been mainstream for morphology analysis. In this thesis, the instance segmentation methods are implemented to segment the outlines of granular particles in packed conditions, and a scale-adaptive training strategy for the Mask R-CNN model is proposed to overcome the limitations of scale problems. The foreground particles and the occluded particles are successfully distinguished and only the identified foreground particles are utilized as the representatives for morphology evaluation. The results demonstrate the feasibility and effectiveness of the proposed method for onsite application in densely packed conditions of granular aggregates. | en_US |
| dcterms.abstract | To avoid the negative impact caused by the occluded particles in 2D space, a framework of SOLO-PCNet that can automatically segment all the particles and predict the complete contours of the occluded particles in the densely packed materials is proposed. Through the proposed framework, the packed particles in images are well segmented and the occluded particles are completed so that the morphology characteristics can be accurately calculated based on the complete outlines of the particles. Regarding of 3D point clouds, a self-supervised transformer-based encoder and decoder model of SSPoinTr-GM is developed for the completion of partially occluded point clouds of granular particles. The corresponding morphology characteristics of the individual particles are explored based on the completed particles and display a high consistency with the ground truth. The framework and the model provide effective methods for fast and accurate evaluation of morphology characteristics of occluded granular particles. | en_US |
| dcterms.abstract | Finally, to dynamically acquire the 3D characteristics of moving granular particles, this thesis proposes a deep-learning-enhanced approach to achieve 3D reconstruction of particles through a dynamic particle tracking algorithm to extract 2D projections. Then a corresponding 3D skeleton network from 2D outlines is generated and the particle surface of individual particles is reconstructed by spatial interpolations so that the corresponding shape indexes are calculated. The results demonstrated that the proposed framework is robust and efficient in reconstructing 3D particles and offers a rapid and low-cost practical solution for convincible numerical studies of granular soils. | en_US |
| dcterms.extent | xviii, 143 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2024 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.LCSH | Granular materials | en_US |
| dcterms.LCSH | Soil mechanics | en_US |
| dcterms.LCSH | Deep learning (Machine learning) | en_US |
| dcterms.LCSH | Computer vision | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13020