A Novel Bilinear Feature and Multi-Layer Fused Convolutional Neural Network for Tactile Shape Recognition

Area of Science:

• Robotics and Machine Perception
• Computer Vision
• Artificial Intelligence

Background:

• Convolutional neural networks (CNNs) show promise for tactile shape recognition by learning from pressure data.
• Low-resolution and blurred tactile images, due to sensor limitations, pose challenges for accurate shape recognition.
• Existing methods struggle with the inherent noise and low fidelity of tactile sensing.

Discussion:

• The proposed bilinear feature and multi-layer fused convolutional neural network (BMF-CNN) addresses limitations of traditional CNNs in tactile image analysis.
• Bilinear feature calculation enhances the network's capability to extract discriminative features from tactile data.
• Multi-layer fusion leverages complementary information across different network depths, improving feature utilization efficiency.

Key Insights:

• BMF-CNN achieves 98.64% average accuracy on a 26-class letter-shape tactile image dataset with complex edges.
• The model demonstrates superior performance compared to traditional CNNs and artificial feature-based methods for tactile shape recognition.
• Effective handling of low-resolution and blurred tactile images is a key advantage of the proposed BMF-CNN.

Outlook:

• Further research can explore BMF-CNN for more complex object recognition tasks in robotics and human-computer interaction.
• Optimization of sensor technology could complement deep learning approaches for even higher tactile recognition fidelity.
• The BMF-CNN framework offers a robust foundation for advancing tactile sensing and interpretation in artificial systems.