基于级联宽度学习的多模态材质识别

PDF下载HTML

Title:

Cascade broad learning for multi-modal material recognition

作者:

王召新¹, 续欣莹¹, 刘华平^2,3, 孙富春^2,3

1. 太原理工大学 电气与动力工程学院，山西 太原 030600;
2. 清华大学 计算机科学与技术系，北京 100084;
3. 清华大学 智能技术系统国家重点实验室，北京 100084

Author(s):

WANG Zhaoxin¹, XU Xinying¹, LIU Huaping^2,3, SUN Fuchun^2,3

1. College of Electrical and Power Engineering, Taiyuan University of Technology, Taiyuan 030600, China;
2. Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China;
3. State Key Laboratory of Intelligent Technology and System, Tsinghua University, Beijing 100084, China

关键词:

级联结构; 宽度学习方法; 多模态融合; 材质识别; 光谱数据; 同构数据; 特征提取; 神经网络

Keywords:

cascade structure; broad learning method; multi-modal fuse; material recognition; spectral data; homogeneous data; feature extraction; neural network

分类号:

TP391

DOI:

10.11992/tis.201908021

摘要:

材质识别在机器人与周围环境的相互作用中起着至关重要的作用，视觉、触觉和听觉模式可以提供不同材质的不同特性，如何利用不同模态的信号快速、高效地完成材质识别任务是亟待解决的问题。并且在现实应用中，传感器收集的数据量不大，无法为深度神经网络提供足够的数据进行学习训练。为此，本文将级联宽度学习这种泛化性能好的算法应用在小样本的材质识别任务上。首先，将两组同构多模态数据进行特征融合，之后使用级联特征节点的宽度学习进行特征学习，最终得到材质分类结果。最后，针对公开数据开展实验评估。结果表明，本文提出的方法与其他算法相比，在完成材质识别任务的同时，降低了训练时间，提高了分类性能。

Abstract:

Material recognition plays a vital role in the interaction betwee-n the robot and the surrounding environment. The visual, tactile and auditory modalities can provide different properties of various materials. How to use signals of different modalities to complete the task of material identification quickly and efficiently is an urgent problem to be solved. Moreover, in practical applications, the data collected by the sensor is limited, so it cannot provide enough data for deep neural network for learning and training. To this end, this paper applies the cascade broad learning with good generalization performance to the material recognition task of small samples. Firstly, feature fusion of two sets of homogeneous multi-modal data is carried out, and then feature learning is carried out by using the broad learning of cascading feature nodes, The results show that compared with other methods, the method proposed in this paper reduces the training time and improves the classification performance while completing the material recognition tasks.