Plantformer: plant point cloud completion based on local–global feature aggregation and spatial context-aware transformer

Plant phenotypic analysis plays a crucial role in plant breeding and transgenic research. Three-dimensional (3D) point cloud is a powerful paradigm in plant phenotypic analysis, which can effectively represent 3D structure and alleviate occlusion-related issues. Moreover, it is imperative to reconst...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Neural computing & applications Ročník 37; číslo 4; s. 2747 - 2762
Hlavní autori: Li, Xiaomeng, Li, Fei, Qi, Yanyu, Li, Zhenbo
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: London Springer London 01.02.2025
Springer Nature B.V
Predmet:
Agriculture
Artificial Intelligence
Computational Biology/Bioinformatics
Computational Science and Engineering
Computer Science
Context
Data Mining and Knowledge Discovery
Datasets
Fruits
Genotype & phenotype
Image Processing and Computer Vision
Image reconstruction
Methods
Occlusion
Original Article
Probability and Statistics in Computer Science
Strawberries
Three dimensional analysis
Three dimensional models
Plant phenotype
Local–global feature aggregation
Spatial context-aware
Three-dimensional point cloud completion
ISSN:0941-0643, 1433-3058
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Plant phenotypic analysis plays a crucial role in plant breeding and transgenic research. Three-dimensional (3D) point cloud is a powerful paradigm in plant phenotypic analysis, which can effectively represent 3D structure and alleviate occlusion-related issues. Moreover, it is imperative to reconstruct incomplete 3D point clouds of plants collected in complex planting scenarios. However, existing methods for plant point cloud completion mainly focus on the extraction of global features and exhibit limitations in effectively aggregating local and global features well, which fails to capture the intricate local geometric structures inherent in plant organs, thereby making it difficult to satisfy completion results. To address these challenges, we proposed a novel fine-grained point cloud completion method, namely, PlantFormer, to generate the complete point cloud from its partial observation. We propose an edge-convolution attention module to aggregate local–global features, which not only captures general geometric structures but also preserves local regional information. Furthermore, a spatial context-aware transformer was introduced to achieve a fine upsample effect on the plant point cloud. More specifically, due to the absence of high-quality datasets, we first conducted PlantCom3D, a plant point cloud completion dataset containing multiple species and growing environments. Extensive experiments demonstrate that our proposed model surpasses comparative models across various metrics. It achieves a C D L 1 of 2.879, with a notable improvement of 12.8 % compared to the previous optimal model.
Bibliografia:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0941-0643
1433-3058
DOI:10.1007/s00521-024-10659-4