利用光纤布拉格光栅在触觉传感方面的高灵敏和柔韧性优势，进行了滑触感知和分类训练研究，实现了在线材质识别。通过理论分析，优化光纤光栅封装，搭建了光纤光栅滑触感知平台，并研究其上位机控制方法及材质在线识别分类算法，提取中心波长的均值最大差值、差值、极差特征作为三维特征，应用支持向量机算法进行分类训练。训练结果表明，在5、10、15 cm/s滑移的混合数据集下，对粗布、PLA、砂纸800目的分类准确度达96.6%。相较于其他特征分类法，具有更好的分类能力和适应不同滑移速度的优势。在5~15 cm/s随机滑速的36次（3类材质×3个样品×4次滑移）验证测试中正确识别了34次。研究成果可为智能感知机器人提供一种在线新颖的材质识别方法。

With the development of smart robots, intelligent tactile sensing is increasingly applied in industrial production, which can greatly improve efficiency and accuracy. Compared with traditional electrical sensors, optical fiber Bragg Grating (FBG) sensors have significant advantages, such as flexibility, electromagnetic immunity, and small size. They also demonstrate high sensitivity and rapid response in perceiving strain and pressure. Current researches on FBG-based tactile sensing mainly focus on strain, temperature, sliding positioning and contact force deduced from the Bragg wavelength shift of FBG. However, there are relatively few researches on combining feature extraction, machine learning, and other cutting-edge technologies to achieve more sophisticated intelligent perception, such as material recognition.In this work, we presented a FBG based sliding-tactile sensing and classification training method for online material recognition by the differential properties of contact surface materials, such as roughness and stick-slip phenomenon. We developed a horizontal two-layer silicone rubber covered FBG sensing unit and its sliding-tactile perception system. When sliding on the certain material, a continuous strain exerts to FBG through the silicone rubber sensing unit and FBG's response changes.To classify efficiently, this paper extracted the mean maximum difference λB(max)-λBˉ, extreme difference λB(max)-λB(min), and standard deviation Δλ(std) of the FBG's wavelengths as the three-dimensional feature for mapping the material properties. And the classification training of the Support Vector Machine (SVM) algorithm and its classification model was developed. The results show that the classification accuracy is 96.6% for rough cloth, PLA and 800-grit sandpaper under the mixed dataset of 5 cm/s, 10 cm/s and 15 cm/s sliding speeds. Compared with the direct wavelength and traditional mean/median feature classification methods, this three-dimensional feature-based method exhibits superior classification capability and adaptability.In order to achieve further intelligent applications, this paper also designs an interactive computer control system, including wavelength acquisition, speed control and material recognition result display. It can control the sliding speed and online material recognition as well. Utilizing the prediction function trainedModel.predictFcn(t_test), the corresponding predicted results were presented after extracting three-dimensional features. In 36 tests, 5~15 cm/s random sliding speed (3 types of materials×3 samples×4 times slip) were carried out, and the correct predictions were 34 tests, which verifies that this method is effective and accurate.This work indicates that the FBG sensor has great potential in the field of material recognition by slip-tactile sensing. The research results can provide a novel online material recognition method for intelligent sensing robots.