开放环境下未知材质的识别技术

针对开放环境下未知物体材质识别的问题，本文提出一种利用欧氏距离区分未知类别和已知类别的物体材质识别方法框架，在该框架下利用支持向量机对物体材质进行识别，分类效果显著。该方法利用距离度量中的欧氏距离与阈值进行比较，距离的均值小于阈值的物体判定为已知类别物体材质，并进行分类；距离大于阈值的物体判定为未知类别物体材质，并利用支持向量机算法进行重新学习识别。本文在慕尼黑工业大学的触觉数据集中的声音数据上进行实验，对比了5种距离度量方法，选择了欧氏距离；与开集稀疏表示分类方法对比，显示出本文提出的方法在声音数据集上具有一定的优势；通过实验选出了合理的阈值，并最终实现了开放环境下识别所有物体材质。实验验证了该框架可以很好地解决开放环境下触觉感知信息的物体材质识别问题。

Recognition of unknown materials in an open environment

Considering the problem of unknown object material recognition in an open environment, this paper proposes an object material recognition method framework that uses the Euclidean distance to distinguish unknown and known categories. Under this framework, a support vector machine is used to recognize object materials, and the classification effect is remarkable. This method uses the Euclidean distance in the distance metric to compare the thresholds. Objects whose average distances are less than the threshold are classified as materials of a known class; objects with distances greater than the threshold are classified as materials of an unknown class and use a support vector machine algorithm for re-learning recognition. Experiments are conducted on sound data in a haptic data set by the Technical University of Munich. Five distance measurement methods are compared, and finally, the Euclidean distance is selected. A comparison with the open set sparse representation classification method shows that the method proposed in this paper has certain advantages on the sound data set. A reasonable threshold is selected through experiments, and finally all object materials are recognized in an open environment. Experiments verify that the framework can solve the problem of object material recognition of tactile perception information in an open environment.