| 摘 要: 触觉传感器采集的一维触觉数据能够用于识别并区分物体的特征,进而实现对物体类别的分类。文章以轻量级卷积神经网络GhostNet为基础框架,提出了一种改进的复合损失函数,以提升模型的分类性能。为进一步适应一维触觉数据的特性,研究对GhostNet模型进行了结构上的改进,使其能够高效处理一维数据。同时,研究将改进后的GhostNet与GRU(GatedRecurrentUnit)网络相结合,构建了GhostNet-GRU (GhostNetwork-Gated RecurrentUnit)网络结构。实验结果表明,采用复合损失函数后,网络精度提高了1.85%,并且与残差网络ResNet相比,网络精度提高了3.42%,证明了所提出改进网络结构的有效性和实用价值。 |
| 关键词: |
|
中图分类号: TP391
文献标识码: A
|
|
| ResearchonOne-Dimensional Signal Recognition of Improved GhostNet Basedon Network Reconstruction |
|
YU Hang, CHEN Yeye, LI Handong
|
(Electrical Engineering College, Guizhou University, Guiyang 550025, China)
2926479961@qq.com; 18892337396@163.com; 470394668@qq.com
|
| Abstract: One-dimensional tactile data collected by tactile sensors can be used to recognize and distinguish the characteristics of objects, thus facilitating the classification of object categories. This paper proposes an improved composite loss function based on the foundational framework of lightweight convolutional neural network GhostNet in order to enhance the model's classification performance. To further adapt to the characteristics of one-dimensional tactile data, structural improvements are made to the GhostNet model, enabling it to efficiently process one-dimensional data. Additionally, the improved GhostNet is combined with a Gated Recurrent Unit (GRU) network to construct the GhostNet-GRU (Ghost Network-Gated Recurrent Unit) architecture. Experimental results indicate that the application of the composite loss function increases the network accuracy by 1.85%, and compared to the Residual Network (ResNet), the accuracy improved by 3. 42%, demonstrating the effectiveness and practical value of the proposed improved network structure. |
| Keywords: |
