Object Recognition Using Shape and Texture Tactile Information: A Fusion Network Based on Data Augmentation and Attention Mechanism

Currently, most tactile-based object recognition algorithms focus on single shape or texture recognition. However, these single attribute-based recognition methods perform poorly when dealing with objects with similar shape or texture characteristics. Research on integrating shape and texture attrib...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:IEEE transactions on haptics Ročník 18; číslo 1; s. 136 - 150
Hlavní autoři: Wang, Bin, Li, Baojiang, Li, Liang, Zhang, Zhekai, Qiu, Shengjie, Wang, Haiyan, Wang, Xichao
Médium: Journal Article
Jazyk:angličtina
Vydáno: United States IEEE 01.01.2025
Témata:
Accuracy
Algorithms
Attention - physiology
Attention mechanisms
attention-based fusion network
Data augmentation
Feature extraction
Humans
Image recognition
Neural Networks, Computer
Object recognition
Pattern Recognition, Automated - methods
Shape
shape and textur attributes
Tactile sensors
Tactile-based object recognition algorithms
Touch
Touch Perception - physiology
Transformers
TSMFormer
ISSN:1939-1412, 2329-4051, 2329-4051
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Currently, most tactile-based object recognition algorithms focus on single shape or texture recognition. However, these single attribute-based recognition methods perform poorly when dealing with objects with similar shape or texture characteristics. Research on integrating shape and texture attributes is still limited, and existing feature fusion mechanisms tend to rely on simple connectivity while ignoring the interactions between different features. To address this issue, we propose a novel attention-based fusion network, TSMFormer, which classifies by integrating shape and texture information and harnesses the global learning capabilities of attention mechanisms to explore interactions between shape and texture in tactile images. Considering the advantages of Transformer networks in handling large datasets, we expanded the existing tactile image dataset through data augmentation. Extensive comparative experiments on this dataset show that the accuracy of the network combining texture and shape information is significantly improved to 99.3%. Comparisons with existing fusion methods further validate the effectiveness of our proposed attention fusion mechanism. The results demonstrate that TSMFormer is highly valuable for research, as it fuses texture and shape information in tactile images through an attention mechanism. Additionally, it shows great potential for practical applications such as robot grasping and automatic quality inspection in industrial environments.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:1939-1412
2329-4051
2329-4051
DOI:10.1109/TOH.2024.3502910